Загрузка видео...

Не удалось загрузить видео

Возникла проблема при загрузке этого видео. Это может быть связано с временными проблемами сети или видео может быть недоступно.

На главную

Someone just used DeepSeek-V4.1-Flash + Hermes Agent to optimize Resident Evil 7 on a OnePlus 12R (Snapdragon 8 Gen 2). The result: 4K textures → 1024/512px Texture data: ~20GB → 8GB Performance: ~20 FPS → stable 30 FPS

智东西China AI News

2,522 subscribers

28,580 просмотров • 10 дней назад •via X (Twitter)

Игры Наука и технологии

Anya Rossi• Live Now

Private livecam show

Комментарии: 3

Фото профиля witcheer

witcheer10 дней назад

yo that's so cool

Фото профиля Pyantler

Pyantler10 дней назад

Goes to show devs are literal retards

Фото профиля CauseWhyNot

CauseWhyNot10 дней назад

💀 this is some crazy use, how was this optimised sooo aggressively?

Похожие видео

I ran a 4K HEVC video export test on both the Dimensity 9500 (vivo X300) and Snapdragon 8 Elite Gen 5 (iQOO 15). I used VN, max performance mode on both. Dimensity 9500: 2 min 7 sec Snapdragon 8 Elite Gen 5: 4 min 17 sec That’s almost 2 × faster export on the Dimensity device.

I ran a 4K HEVC video export test on both the Dimensity 9500 (vivo X300) and Snapdragon 8 Elite Gen 5 (iQOO 15). I used VN, max performance mode on both. Dimensity 9500: 2 min 7 sec Snapdragon 8 Elite Gen 5: 4 min 17 sec That’s almost 2 × faster export on the Dimensity device.

Rishabh Negi

19,789 просмотров • 6 месяцев назад

Borderlands 4 Performance Summary in 2 minutes: 0:00 - Native Res Badass Settings the game runs bad 0:30 - Upscaling Performance Mode Badass Settings, the game barely manages to hit 60FPS 1:20 - Frame Gen turned on and now you have 100 FPS Some people will turn all these settings on and be happy with it but in my honest opinion these are used as crutches to make the game look as if it performs good. Borderlands 4 has a certain art style that doesn't look like it needs these kind of settings to be turned on to play at a high framerate but sadly it does and you can defend this if you want but the future of gaming looks bleak, using this tech to not optimize the games further will just result in input latency increase as every game will be delivered with frame-gen turned on in the devs mind.

Borderlands 4 Performance Summary in 2 minutes: 0:00 - Native Res Badass Settings the game runs bad 0:30 - Upscaling Performance Mode Badass Settings, the game barely manages to hit 60FPS 1:20 - Frame Gen turned on and now you have 100 FPS Some people will turn all these settings on and be happy with it but in my honest opinion these are used as crutches to make the game look as if it performs good. Borderlands 4 has a certain art style that doesn't look like it needs these kind of settings to be turned on to play at a high framerate but sadly it does and you can defend this if you want but the future of gaming looks bleak, using this tech to not optimize the games further will just result in input latency increase as every game will be delivered with frame-gen turned on in the devs mind.

NikTek

359,686 просмотров • 1 год назад

#Helldivers2, a #PlayStation first-party title, runs smoother on #Xbox than on PS5. Let that sink in 🔥😘 Digital Foundry breaks it down: “FlopDivers 2” performs better on Xbox across the board. ✅ In Performance Mode, Xbox flexes its superior VRR 🦾 for a smoother, more stable experience. The FPS drops outside of the VRR window of PS5. PS5 also waste developers time and holds the gen back for lacking important features. ✅ In 30 FPS Quality Mode, Xbox delivers perfect frame pacing unlike PS5, which still struggles with judder 🥲 Why do you think #Xbox systems manage to outperform #PlayStation in their own games? #Xbox #XboxGamePass

#Helldivers2, a #PlayStation first-party title, runs smoother on #Xbox than on PS5. Let that sink in 🔥😘 Digital Foundry breaks it down: “FlopDivers 2” performs better on Xbox across the board. ✅ In Performance Mode, Xbox flexes its superior VRR 🦾 for a smoother, more stable experience. The FPS drops outside of the VRR window of PS5. PS5 also waste developers time and holds the gen back for lacking important features. ✅ In 30 FPS Quality Mode, Xbox delivers perfect frame pacing unlike PS5, which still struggles with judder 🥲 Why do you think #Xbox systems manage to outperform #PlayStation in their own games? #Xbox #XboxGamePass

AnaAnsan (XBOX Ultimate Fan)

27,913 просмотров • 1 год назад

A CHINESE LAB VISUALIZED 2,000 NEURONS AND 1,191,000 SYNAPSES IN 3D. HERMES AGENT HIT 140,000 STARS RUNNING THE SAME NETWORK LOCALLY FOR $0 hermes is the first agent that writes its own skills from experience. complete a task once and it saves the procedure as a markdown file for next time agents with 20+ self-created skills complete similar tasks 40% faster than fresh instances. less time and less tokens to get the same result qwen 3.6 35b outperforms last year's 120b models and runs on 20gb of memory. the intelligence that needed a data center now fits on your desk setup takes 30 minutes. install lm studio, pull qwen 3.6, install hermes, point it at localhost. zero api fees, zero data leaving your machine most people pay for cloud agents that forget everything between sessions. the ones running hermes locally in 2026 will look very far ahead in 2028 bookmark this and read the article below

A CHINESE LAB VISUALIZED 2,000 NEURONS AND 1,191,000 SYNAPSES IN 3D. HERMES AGENT HIT 140,000 STARS RUNNING THE SAME NETWORK LOCALLY FOR $0 hermes is the first agent that writes its own skills from experience. complete a task once and it saves the procedure as a markdown file for next time agents with 20+ self-created skills complete similar tasks 40% faster than fresh instances. less time and less tokens to get the same result qwen 3.6 35b outperforms last year's 120b models and runs on 20gb of memory. the intelligence that needed a data center now fits on your desk setup takes 30 minutes. install lm studio, pull qwen 3.6, install hermes, point it at localhost. zero api fees, zero data leaving your machine most people pay for cloud agents that forget everything between sessions. the ones running hermes locally in 2026 will look very far ahead in 2028 bookmark this and read the article below

starmex

13,420 просмотров • 3 месяцев назад

HERMES AGENT NOW RUNS ON AN 8GB LAPTOP GPU JUST AS EASILY AS IT RUNS ON A 128GB MINI PC Nous Research shipped the official Hermes Agent Desktop App this week. Someone pointed it at a local llama server running on an RTX 4060 with 16GB system RAM. The integration took two minutes The model behind it: Gemma 4 26B MoE, QAT quantized, running on 8GB of VRAM. A 60k token prompt held a stable 20 tokens a second, flat, no slowdown as context grew. The flags were nothing exotic, just -cmoe -c 248000 on llama.cpp What that 8GB setup does out of the box: reads and patches its own code, runs it in a terminal, debugs errors, manages GitHub repos, spawns sub-agents for parallel work. Browses the web with vision to debug a UI. Schedules cron jobs in plain language. Connects to Notion, Google Workspace, Linear, and Obsidian to manage tasks on its own That's the same agent layer running on a Minisforum MS-S1 MAX with 128GB of unified memory, 96GB of it to the GPU, holding a 120B model at 56 tokens a second instead of a 26B model at 20. Same software, same tool execution, same zero API key. The only thing that changes between an $800 laptop and a $2,000 mini PC is how big a model you can afford to run underneath it The barrier to running a real autonomous agent locally didn't just drop. It dropped all the way down to hardware most people already own

HERMES AGENT NOW RUNS ON AN 8GB LAPTOP GPU JUST AS EASILY AS IT RUNS ON A 128GB MINI PC Nous Research shipped the official Hermes Agent Desktop App this week. Someone pointed it at a local llama server running on an RTX 4060 with 16GB system RAM. The integration took two minutes The model behind it: Gemma 4 26B MoE, QAT quantized, running on 8GB of VRAM. A 60k token prompt held a stable 20 tokens a second, flat, no slowdown as context grew. The flags were nothing exotic, just -cmoe -c 248000 on llama.cpp What that 8GB setup does out of the box: reads and patches its own code, runs it in a terminal, debugs errors, manages GitHub repos, spawns sub-agents for parallel work. Browses the web with vision to debug a UI. Schedules cron jobs in plain language. Connects to Notion, Google Workspace, Linear, and Obsidian to manage tasks on its own That's the same agent layer running on a Minisforum MS-S1 MAX with 128GB of unified memory, 96GB of it to the GPU, holding a 120B model at 56 tokens a second instead of a 26B model at 20. Same software, same tool execution, same zero API key. The only thing that changes between an $800 laptop and a $2,000 mini PC is how big a model you can afford to run underneath it The barrier to running a real autonomous agent locally didn't just drop. It dropped all the way down to hardware most people already own

NO1ennn

40,079 просмотров • 3 месяцев назад

HERMES AGENT CAN RUN YOUR SEO. CONNECT IT TO GOOGLE SEARCH CONSOLE AND GOOGLE ANALYTICS. IT MONITORS, REPORTS, AND WRITES CONTENT BASED ON YOUR ACTUAL DATA. stop paying an SEO agency. stop doing the tedious work yourself. Hermes handles it 24/7. WHAT THE SEO AGENT DOES: → pulls clicks, impressions, CTR, and position data from Google Search Console automatically → tracks traffic, user behavior, and conversions from Google Analytics → checks which pages are indexed and which are not → submits sitemaps for indexing → inspects URLs for crawl or indexing issues → identifies ranking drops and keyword opportunities → writes content based on what your data says works → generates weekly SEO performance reports → delivers everything to Telegram CONNECT GOOGLE SEARCH CONSOLE: two paths: 1. COMPOSIO (managed, easiest): paste this into Hermes chat: https:// composio. dev/hermes or add to config.yaml: mcp_servers: composio: url: "https:// connect.composio. dev /mcp" headers: x-consumer-api-key: "YOUR_COMPOSIO_API_KEY" Hermes prompts you to authenticate. one OAuth flow. done. 2. CLAWLINK (one-click): 9 Google Search Console tools exposed via MCP. hosted auth. nothing to run or maintain. paste the install prompt into Hermes chat. CONNECT GOOGLE ANALYTICS: same Composio setup. one MCP endpoint handles both Search Console and Analytics. authenticate once. both data sources available. your agent can now query: → search analytics (clicks, impressions, CTR, position) → traffic by source and landing page → user behavior and conversions → indexing status for any URL → sitemap status WHAT TO AUTOMATE WITH CRON: weekly SEO report (Monday 8am): "pull search analytics for last 7 days. compare vs previous week. flag any keyword that dropped more than 5 positions. flag any page that lost more than 20% clicks. deliver report to Telegram." daily indexing check (6am): "check if any new pages are not indexed. if found, submit sitemap and report to Telegram." wakeAgent gate: skip if all pages indexed. content opportunity scan (weekly): "find queries where my site appears on page 2 (positions 11-20) with high impressions. these are the keywords one good article could push to page 1. deliver list to Telegram with suggested topics." CONTENT WRITING FROM YOUR DATA: the difference between generic SEO content and content that ranks: your agent has your Search Console data. "write a blog post targeting [keyword]. my current position is 14 with 2,400 monthly impressions. check what pages currently rank 1-3 for this keyword. write something better. include the gaps they miss." the agent researches competitors via Firecrawl, checks your existing content in the wiki, and drafts based on real data. not guesswork. WHAT THIS REPLACES: → SEO agency: $1,000-5,000/month → SEO tool subscriptions: $100-300/month → manual reporting: 3-5 hours/week → manual content research: 2-4 hours/week Hermes SEO agent: one profile with two MCPs. cron jobs handle the monitoring. you handle the decisions. SETUP IN 10 MINUTES: 1. create a profile: hermes profile create seo-agent 2. write SOUL.md: "you are an SEO specialist. monitor search performance daily. flag ranking drops and opportunities. write content based on Search Console data. weekly report every Monday." 3. connect Google Search Console + Analytics via Composio or ClawLink 4. set cron jobs (weekly report, daily index check, content opportunity scan) 5. set model: DeepSeek V4 for routine monitoring. Sonnet for content writing. 6. connect to Telegram for delivery. the agent runs. you review reports. rankings improve because you stopped guessing and started using your own data. comment HERMES and I'll send you the full setup guide for running Hermes Agent as your SEO specialist. full Hermes architecture deep-dive in the article 👇

YanXbt

41,034 просмотров • 2 месяцев назад

gemini 3.7 flash vs deepseek v4 pro 0813 vs muse spark 1.2 – on voxel city dioramas three models each built three crossy road-style 3d scenes – a construction site, a nyc intersection, a river with a drawbridge – as single self-contained html files the setup: Nous Research's hermes agent cli on OpenRouter, three.js skills preloaded, identical prompts per scene tasks: 1. construction site – tower crane on a working lift loop, paver laying fresh road, roller compacting it behind 2. nyc crossing – four-way intersection with a traffic light state machine, queuing cars, pedestrians crossing on the walk signal 3. river drawbridge – double-leaf bascule that lifts for tall boats, cars queuing at the barriers, animated water every scene: Three.js r185, box geometry only, a locked 20-color palette, four camera presets, and a day/night mode with bloom. one file, no build step, no assets models: Google DeepMind gemini 3.7 flash, DeepSeek v4 pro 0813, AI at Meta muse spark 1.2 muse and gemini finished every scene in two to three minutes. deepseek took 15 to 41 minutes per scene - build time, all three scenes #1 gemini 3.7 flash – 6m 43s #2 muse spark 1.2 – 7m 20s #3 deepseek v4 pro – 91m 25s - total tokens #1 muse spark 1.2 – 440,279 #2 gemini 3.7 flash – 713,855 #3 deepseek v4 pro – 20,957,568 - total price #1 muse spark 1.2 – $0.53 #2 gemini 3.7 flash – $0.56 #3 deepseek v4 pro – $4.57 - agent calls across the three builds #1 muse spark 1.2 – 12 #2 gemini 3.7 flash – 18 #3 deepseek v4 pro – 143 observations: • muse won two of the three scenes on looks with the smallest files in the test – 887 to 1,042 lines against gemini's 1,934 to 2,377. cheapest, fastest to a good frame, and shortest turned out to be the same column • deepseek burned 20.96m tokens – 29x gemini, 48x muse – across 143 agent calls. prompt caching is the only reason that cost $4.57: the cache discount absorbed roughly $30 of resent context • gemini was the only model whose files needed zero fixes to render – and the only one whose night mode is cosmetic. the sky never darkens and one camera button does nothing. clean code for a scene it never looked at follow thehype. for 24/7 ai news, analysis and breakdowns

gemini 3.7 flash vs deepseek v4 pro 0813 vs muse spark 1.2 – on voxel city dioramas three models each built three crossy road-style 3d scenes – a construction site, a nyc intersection, a river with a drawbridge – as single self-contained html files the setup: Nous Research's hermes agent cli on OpenRouter, three.js skills preloaded, identical prompts per scene tasks: 1. construction site – tower crane on a working lift loop, paver laying fresh road, roller compacting it behind 2. nyc crossing – four-way intersection with a traffic light state machine, queuing cars, pedestrians crossing on the walk signal 3. river drawbridge – double-leaf bascule that lifts for tall boats, cars queuing at the barriers, animated water every scene: Three.js r185, box geometry only, a locked 20-color palette, four camera presets, and a day/night mode with bloom. one file, no build step, no assets models: Google DeepMind gemini 3.7 flash, DeepSeek v4 pro 0813, AI at Meta muse spark 1.2 muse and gemini finished every scene in two to three minutes. deepseek took 15 to 41 minutes per scene - build time, all three scenes #1 gemini 3.7 flash – 6m 43s #2 muse spark 1.2 – 7m 20s #3 deepseek v4 pro – 91m 25s - total tokens #1 muse spark 1.2 – 440,279 #2 gemini 3.7 flash – 713,855 #3 deepseek v4 pro – 20,957,568 - total price #1 muse spark 1.2 – $0.53 #2 gemini 3.7 flash – $0.56 #3 deepseek v4 pro – $4.57 - agent calls across the three builds #1 muse spark 1.2 – 12 #2 gemini 3.7 flash – 18 #3 deepseek v4 pro – 143 observations: • muse won two of the three scenes on looks with the smallest files in the test – 887 to 1,042 lines against gemini's 1,934 to 2,377. cheapest, fastest to a good frame, and shortest turned out to be the same column • deepseek burned 20.96m tokens – 29x gemini, 48x muse – across 143 agent calls. prompt caching is the only reason that cost $4.57: the cache discount absorbed roughly $30 of resent context • gemini was the only model whose files needed zero fixes to render – and the only one whose night mode is cosmetic. the sky never darkens and one camera button does nothing. clean code for a scene it never looked at follow thehype. for 24/7 ai news, analysis and breakdowns

thehype.

29,033 просмотров • 1 месяц назад

union alpha (unbiased pareto) vs deepseek v4.1 flash vs muse spark 1.3 – three paintings in three.js the setup: one four-line prompt plus the painting as an image, through OpenRouter. no agent loop, no renders, no feedback – the model writes one html file blind and we open it. Three.js from a cdn, every texture generated in code. when a provider cut the stream early we sent the partial back and said continue exactly where you stopped tasks: 1. the starry night – van gogh, 1889 2. the persistence of memory – dalí, 1931 3. poppies at argenteuil – monet, 1873 two rules in every brief: keep the painting's palette, brushwork and mood, and reply with the code only models: Unbiased AI union alpha (stealth, free), DeepSeek deepseek v4.1 flash, AI at Meta muse spark 1.3 total cost, three scenes #1 union alpha – free (list price: $1.04) #2 muse spark 1.3 – $0.117 #3 deepseek v4.1 flash – $0.162 generation time, three scenes #1 muse spark 1.3 – 4m 45s #2 deepseek v4.1 flash – 11m 30s #3 union alpha – 29m 13s total completion tokens #1 muse spark 1.3 – 26,502 #2 union alpha – 123,756 #3 deepseek v4.1 flash – 156,028 lines of code shipped #1 muse spark 1.3 – 862 #2 union alpha – 1,715 #3 deepseek v4.1 flash – 2,816 observations: • union alpha reads the painting like an art historian. it named every work unprompted, then broke each into parts: dalí's watches deformed along a bezier curve, monet's poppies as instanced brush dabs under a wind shader. no other model went that deep on a one-shot • it is the only model that made the paintings move the way they were painted. in the monet, the woman and the child walk the field on catmull-rom paths, pollen drifts, poppies are brush dabs in a point shader that sway in the wind. deepseek and muse left the figures standing • its dalí is an inventory of the canvas: three soft clocks draped along one parametric curve, a drip falling off the hanging one, a fly, ants on the pocket watch as an instanced mesh. 17 named parts in all. nobody else drew the drip • its starry night is shader work end to end: shared glsl noise, a vortex field for the sky, billboarded shader quads for the moon and stars, a painterly surface shader for the hills, and windows that flicker on their own timers. the file reads like a demoscene entry, not a model output what union alpha is: • we asked it. the stealth window had closed a day after launch and the api answered: "this model was unbiased's pareto". pareto is from circuit & chisel, an ex-stripe team that raised $19.2m in sep 2025 per fortune, and now sells "frontier intelligence for 75% less" • pareto is not one model. per unbiased's site it "runs a mix of frontier and open source models against each other on every request" and keeps the best answer. that is the 300-second first token, the tokenizer listed as "other", and the missing reasoning field – a race, not a model • listed at $2.50 in and $7.50 out per million. our three scenes would have cost $1.04 – 6.4x deepseek, 8.9x muse. asked for its cutoff, it dated nothing past may 2025 follow thehype. for 24/7 ai news, analysis and breakdowns

union alpha (unbiased pareto) vs deepseek v4.1 flash vs muse spark 1.3 – three paintings in three.js the setup: one four-line prompt plus the painting as an image, through OpenRouter. no agent loop, no renders, no feedback – the model writes one html file blind and we open it. Three.js from a cdn, every texture generated in code. when a provider cut the stream early we sent the partial back and said continue exactly where you stopped tasks: 1. the starry night – van gogh, 1889 2. the persistence of memory – dalí, 1931 3. poppies at argenteuil – monet, 1873 two rules in every brief: keep the painting's palette, brushwork and mood, and reply with the code only models: Unbiased AI union alpha (stealth, free), DeepSeek deepseek v4.1 flash, AI at Meta muse spark 1.3 total cost, three scenes #1 union alpha – free (list price: $1.04) #2 muse spark 1.3 – $0.117 #3 deepseek v4.1 flash – $0.162 generation time, three scenes #1 muse spark 1.3 – 4m 45s #2 deepseek v4.1 flash – 11m 30s #3 union alpha – 29m 13s total completion tokens #1 muse spark 1.3 – 26,502 #2 union alpha – 123,756 #3 deepseek v4.1 flash – 156,028 lines of code shipped #1 muse spark 1.3 – 862 #2 union alpha – 1,715 #3 deepseek v4.1 flash – 2,816 observations: • union alpha reads the painting like an art historian. it named every work unprompted, then broke each into parts: dalí's watches deformed along a bezier curve, monet's poppies as instanced brush dabs under a wind shader. no other model went that deep on a one-shot • it is the only model that made the paintings move the way they were painted. in the monet, the woman and the child walk the field on catmull-rom paths, pollen drifts, poppies are brush dabs in a point shader that sway in the wind. deepseek and muse left the figures standing • its dalí is an inventory of the canvas: three soft clocks draped along one parametric curve, a drip falling off the hanging one, a fly, ants on the pocket watch as an instanced mesh. 17 named parts in all. nobody else drew the drip • its starry night is shader work end to end: shared glsl noise, a vortex field for the sky, billboarded shader quads for the moon and stars, a painterly surface shader for the hills, and windows that flicker on their own timers. the file reads like a demoscene entry, not a model output what union alpha is: • we asked it. the stealth window had closed a day after launch and the api answered: "this model was unbiased's pareto". pareto is from circuit & chisel, an ex-stripe team that raised $19.2m in sep 2025 per fortune, and now sells "frontier intelligence for 75% less" • pareto is not one model. per unbiased's site it "runs a mix of frontier and open source models against each other on every request" and keeps the best answer. that is the 300-second first token, the tokenizer listed as "other", and the missing reasoning field – a race, not a model • listed at $2.50 in and $7.50 out per million. our three scenes would have cost $1.04 – 6.4x deepseek, 8.9x muse. asked for its cutoff, it dated nothing past may 2025 follow thehype. for 24/7 ai news, analysis and breakdowns

thehype.

23,145 просмотров • 9 дней назад

Playing Bayonetta 3 from scratch after 3 years feels strangely nostalgic and sad at the same time. We had so many great memories with this game (300 hours maybe), but at the same time, it received so much hate from others. Playing it on Switch 2 makes me realize this is probably what we should have had on the original Switch: a beautiful, stable 60 FPS experience. PlatinumGames clearly had to cut the game down to the bone just to make it run. Textures, effects, distant scenery… you can see the compromises everywhere. And that poor Switch used to get so hot it felt like you could cook a chicken on it. Yet the characters and combat still look incredible. Now, on Switch 2, the game runs effortlessly. No heat, no noise, no struggle. Just silent. It honestly feels bittersweet. I’ll keep saying it: we deserve a Bayonetta 3 Nintendo Switch 2 Edition.

Playing Bayonetta 3 from scratch after 3 years feels strangely nostalgic and sad at the same time. We had so many great memories with this game (300 hours maybe), but at the same time, it received so much hate from others. Playing it on Switch 2 makes me realize this is probably what we should have had on the original Switch: a beautiful, stable 60 FPS experience. PlatinumGames clearly had to cut the game down to the bone just to make it run. Textures, effects, distant scenery… you can see the compromises everywhere. And that poor Switch used to get so hot it felt like you could cook a chicken on it. Yet the characters and combat still look incredible. Now, on Switch 2, the game runs effortlessly. No heat, no noise, no struggle. Just silent. It honestly feels bittersweet. I’ll keep saying it: we deserve a Bayonetta 3 Nintendo Switch 2 Edition.

Bayonetta Stuff

33,721 просмотров • 1 месяц назад

$I just crammed the updated Gemma 4 26B A4B QAT (MoE) with 180k context into an 8GB RTX 4060 (8 GB VRAM + 16 GB RAM only!!) and optimized the batch size. 23 tokens/sec decode, 300 tokens/sec prefill Yesterday I showed you a Gemma 4 31B dense model running flawlessly on an RTX 4090. Today, we're breaking the VRAM bank on a budget card using Unsloth’s new Gemma 4 26B (A4B) QAT quants. Following Google’s chat template update that boosted agentic benchmarks by +10%, I pushed this model to its absolute limits. Here is how you squeeze 250k context out of 8GB of VRAM. # The Setup & The Optimization - Hardware: Nvidia RTX 4060 (8GB VRAM) + 16GB System RAM - Environment: CUDA 13.0 build of llama.cpp - Model: gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf - Prompt: 28,000 tokens of prompt for each run If you read my L2 cache breakdown (attached in replies), you know the 4060’s 24MB cache maxes out at `-b 1024 -ub 1024`. Push past that, and prefill crashes. I locked those flags in for every test below to ensure maximum GEMM throughput. # 1. The Raw Context Push (Unquantized KV Cache) First, I wanted to see how far pure 8GB VRAM + 16GB RAM could stretch without touching the KV cache: - 80k Context: Prefill 385 t/s | Decode 25.5 t/s - 120k Context: Prefill 270 t/s | Decode 24 t/s llama.cpp flags: .\llama-server -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -c 120000 --port 8080 -ub 1024 -b 1024 Without KV quantization, 120k is your hard ceiling. push past that prefill throughput drops off a cliff, making the model practically unusable for large agentic workloads. # 2. The Q8 KV Cache Lifeline To survive 250k context on a budget card, you have to quantize the KV cache. I enabled 8 bit KV cache (`-ctk q8_0 -ctv q8_0`) and re ran: - 180k Context: Prefill 280 t/s | Decode 22.8 t/s - 250k Context: Prefill 115 t/s | Decode 20 t/s llama.cpp flags: .\llama-server -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -c 180000 --port 8080 -b 1024 -ub 1024 -ctk q8_0 -ctv q8_0 Result: Q8 KV cache brings 250k context back from the dead. Decode speed stabilizes at a highly usable 20 t/s. You are trading a very small bit amount of reasoning precision for an extra 130,000 tokens of context window. if you own a single rtx 3050, 3060, 3070, 4050, 4060, 5050 or 5060, you must try this model and optimize your batch size for higher prefill. Hugging Face links to the updated Unsloth's QAT quants and performance graph are in the replies below. What model are you running on your 6GB, 8GB or 12GB cards right now? Let's see your setups.$

I just crammed the updated Gemma 4 26B A4B QAT (MoE) with 180k context into an 8GB RTX 4060 (8 GB VRAM + 16 GB RAM only!!) and optimized the batch size. 23 tokens/sec decode, 300 tokens/sec prefill Yesterday I showed you a Gemma 4 31B dense model running flawlessly on an RTX 4090. Today, we're breaking the VRAM bank on a budget card using Unsloth’s new Gemma 4 26B (A4B) QAT quants. Following Google’s chat template update that boosted agentic benchmarks by +10%, I pushed this model to its absolute limits. Here is how you squeeze 250k context out of 8GB of VRAM. # The Setup & The Optimization - Hardware: Nvidia RTX 4060 (8GB VRAM) + 16GB System RAM - Environment: CUDA 13.0 build of llama.cpp - Model: gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf - Prompt: 28,000 tokens of prompt for each run If you read my L2 cache breakdown (attached in replies), you know the 4060’s 24MB cache maxes out at `-b 1024 -ub 1024`. Push past that, and prefill crashes. I locked those flags in for every test below to ensure maximum GEMM throughput. # 1. The Raw Context Push (Unquantized KV Cache) First, I wanted to see how far pure 8GB VRAM + 16GB RAM could stretch without touching the KV cache: - 80k Context: Prefill 385 t/s | Decode 25.5 t/s - 120k Context: Prefill 270 t/s | Decode 24 t/s llama.cpp flags: .\llama-server -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -c 120000 --port 8080 -ub 1024 -b 1024 Without KV quantization, 120k is your hard ceiling. push past that prefill throughput drops off a cliff, making the model practically unusable for large agentic workloads. # 2. The Q8 KV Cache Lifeline To survive 250k context on a budget card, you have to quantize the KV cache. I enabled 8 bit KV cache (`-ctk q8_0 -ctv q8_0`) and re ran: - 180k Context: Prefill 280 t/s | Decode 22.8 t/s - 250k Context: Prefill 115 t/s | Decode 20 t/s llama.cpp flags: .\llama-server -m gemma-4-26B-A4B-it-qat-UD-Q4_K_XL.gguf -c 180000 --port 8080 -b 1024 -ub 1024 -ctk q8_0 -ctv q8_0 Result: Q8 KV cache brings 250k context back from the dead. Decode speed stabilizes at a highly usable 20 t/s. You are trading a very small bit amount of reasoning precision for an extra 130,000 tokens of context window. if you own a single rtx 3050, 3060, 3070, 4050, 4060, 5050 or 5060, you must try this model and optimize your batch size for higher prefill. Hugging Face links to the updated Unsloth's QAT quants and performance graph are in the replies below. What model are you running on your 6GB, 8GB or 12GB cards right now? Let's see your setups.

Alok

36,617 просмотров • 2 месяцев назад

HERMES AGENT IS NOW IN THE CLOUD. NO VPS. NO TERMINAL. NO SETUP. PICK A MODEL. PICK A SERVER SIZE. AGENT IS LIVE IN 60 SECONDS. Nous Portal just launched hosted Hermes Agent. two clicks. one minute. done. Nous Research WHAT THIS MEANS: before today: install Hermes on a VPS or your laptop. configure providers. set up gateway. manage updates. run hermes setup. edit config.yaml. great for power users. friction for everyone else. now: go to pick a model. pick a server size. your agent is live and reachable in 60 seconds. no terminal. no SSH. no Docker. same Hermes. same features. same tools. someone else handles the infrastructure. FOR TEAMS: this is where it gets interesting. spin up agents for everyone at your org. each team member gets their own Hermes instance. granular access controls per user. unified billing through Nous Portal. your team gets Hermes on day one. no DevOps needed. no VPS per person. one admin dashboard. one bill. WHAT'S INCLUDED: → 300+ models via Nous Portal (Claude, GPT, Gemini, DeepSeek, Grok, MiniMax, and more) → Tool Gateway (web search, image generation, TTS, browser automation) → all messaging platforms (Telegram, Discord, Slack, WhatsApp, Signal) → full feature set (profiles, cron, kanban, skills, memory, sub-agents, MoA, /goal, /learn, /journey) → automatic updates ONE PORTAL. FOUR TIERS: Free: $0/month. pay-as-you-go credits from $10. Plus: $20/month. $22 in monthly usage credit. Super: $100/month. $110 in monthly credit. Ultra: $200/month. $220 in monthly credit. highest rate limits. every paid tier includes Tool Gateway. one OAuth. one subscription. no extra API keys. SELF-HOSTED IS NOT GOING ANYWHERE: Hermes is MIT licensed. open source. free forever. you can still run it on your laptop, VPS, or GPU cluster. nothing changes for self-hosted users. the cloud version is for people who want the agent running without managing the machine. pick your path: → self-hosted: full control. you manage everything. → cloud: zero ops. Nous manages infrastructure. → hybrid: self-host your main agent, cloud for team members. HOW TO START: cloud: self-hosted: hermes setup --portal both connect to the same Nous Portal. same models. same tools. same billing. learn how to replace your entire team with 8 hermes agents 👇

HERMES AGENT IS NOW IN THE CLOUD. NO VPS. NO TERMINAL. NO SETUP. PICK A MODEL. PICK A SERVER SIZE. AGENT IS LIVE IN 60 SECONDS. Nous Portal just launched hosted Hermes Agent. two clicks. one minute. done. Nous Research WHAT THIS MEANS: before today: install Hermes on a VPS or your laptop. configure providers. set up gateway. manage updates. run hermes setup. edit config.yaml. great for power users. friction for everyone else. now: go to pick a model. pick a server size. your agent is live and reachable in 60 seconds. no terminal. no SSH. no Docker. same Hermes. same features. same tools. someone else handles the infrastructure. FOR TEAMS: this is where it gets interesting. spin up agents for everyone at your org. each team member gets their own Hermes instance. granular access controls per user. unified billing through Nous Portal. your team gets Hermes on day one. no DevOps needed. no VPS per person. one admin dashboard. one bill. WHAT'S INCLUDED: → 300+ models via Nous Portal (Claude, GPT, Gemini, DeepSeek, Grok, MiniMax, and more) → Tool Gateway (web search, image generation, TTS, browser automation) → all messaging platforms (Telegram, Discord, Slack, WhatsApp, Signal) → full feature set (profiles, cron, kanban, skills, memory, sub-agents, MoA, /goal, /learn, /journey) → automatic updates ONE PORTAL. FOUR TIERS: Free: $0/month. pay-as-you-go credits from $10. Plus: $20/month. $22 in monthly usage credit. Super: $100/month. $110 in monthly credit. Ultra: $200/month. $220 in monthly credit. highest rate limits. every paid tier includes Tool Gateway. one OAuth. one subscription. no extra API keys. SELF-HOSTED IS NOT GOING ANYWHERE: Hermes is MIT licensed. open source. free forever. you can still run it on your laptop, VPS, or GPU cluster. nothing changes for self-hosted users. the cloud version is for people who want the agent running without managing the machine. pick your path: → self-hosted: full control. you manage everything. → cloud: zero ops. Nous manages infrastructure. → hybrid: self-host your main agent, cloud for team members. HOW TO START: cloud: self-hosted: hermes setup --portal both connect to the same Nous Portal. same models. same tools. same billing. learn how to replace your entire team with 8 hermes agents 👇

YanXbt

45,446 просмотров • 2 месяцев назад

HERMES AGENT SUPPORTS 300+ MODELS. PICKING THE RIGHT ONE PER TASK IS THE DIFFERENCE BETWEEN $5/MONTH AND $50. STARTING OUT: Claude Sonnet 4.6. official recommendation from Nous Research. "the model this project was built and tested with." strong reasoning. reliable tool calling. mid-range pricing. PREMIUM TIER: Claude Opus 4.8. best coding benchmarks available. self-correcting reasoning. catches its own mistakes. 1M context. use for demanding tasks where quality matters. GPT-5.5. #1 Chatbot Arena. #1 GPQA Diamond reasoning (94.1%). #1 creative writing. 2M context. handles entire codebases in one pass. Grok 4.30. the only frontier model with live X firehose access. real-time social data, breaking news, market sentiment. connects via Grok OAuth. no separate API key. Grok-Composer-2.5-Fast (v0.17.0). Cursor's coding model. 200K context. available through your Grok subscription via OAuth. no extra cost if you already pay for Grok. MID-RANGE TIER: Claude Sonnet 4.6. best balance of quality and cost for daily use. strongest prose and tool calling in this tier. Gemini 2.5 Pro. Google Search grounding built in. cites sources. verifies claims. pulls current data. 2M context. best for research-heavy workflows. GPT-4.1. reliable tool calling. solid general reasoning. good middle ground when you need OpenAI compatibility. BUDGET TIER: Claude Haiku 4.5. fastest Anthropic model. cheapest paid Claude option. strong at classification, routing, simple queries. use for auxiliary tasks: compression, vision, web extraction, approval scoring. DeepSeek V4. best cost-to-quality ratio in the market. 90% cache discount on repeated context. use for sub-agents and bulk parallel work. DeepSeek V4 Flash. cheapest paid model worth using. 1M context. MIT license. self-hostable. use for cron jobs, monitoring, routine searches. MiniMax M3. Nous Research and MiniMax collaborating on optimization. 1M context via lightning attention. 59% SWE-Bench Pro. beats several premium models on coding. one of the most-used models inside Hermes. FREE / LOCAL: Qwen 3.5 27B via Ollama. 16GB VRAM. reliable tool calling. best free local model for Hermes as of mid-2026. Qwen 3 8B. 8GB VRAM. fits a $7 VPS. handles routine tasks at zero API cost. Llama 4 Maverick. best open-weight tool calling. 1M context. needs more VRAM but strongest local option. HOW TO ASSIGN MODELS: main model: Desktop app / Dashboard → Models → switch sub-agent model: set in Desktop app, Dashboard, or config.yaml: delegation: model: "deepseek/deepseek-v4" auxiliary models (compression, vision, web extract): Desktop app / Dashboard → Models → Auxiliary Haiku 4.5 or Gemini Flash work well here. saves significantly when your main model is premium. per-profile: each Hermes profile gets its own model. Scout on DeepSeek. Analyst on Sonnet. Briefer on budget model. Coder on Opus. per-cron-job: pin a specific model to any cron job. morning brief on Haiku. deep research on Sonnet. monitoring on DeepSeek Flash. each job uses only the model it needs. per-session: /model deepseek/deepseek-v4-flash hot-swap mid-conversation. no restart needed. FALLBACK CHAINS: if your primary model is unavailable, Hermes automatically switches to the next provider. rate limit or server error = next model in the chain. no failed runs. no manual intervention. set in Desktop app, Dashboard, or config.yaml: fallback_providers: - openrouter - nous - codex PROVIDER PATHS: OPENROUTER: 300+ models under one API key. pay per token. most flexible. NOUS PORTAL: 300+ models + Tool Gateway (web search, image gen, TTS, browser). one OAuth. one subscription. 10% off token-billed providers. CHATGPT SUB: GPT-5.5 + Grok via OAuth. included tokens with $20 subscription. OLLAMA: free. local. private. zero API cost. your hardware only. mix providers across profiles and tasks. Scout on OpenRouter. Analyst on Nous Portal. Coder on ChatGPT sub. Monitor on Ollama. THE RULE: premium for work that needs deep reasoning. mid-range for daily driver tasks. budget for volume and background work. free for monitoring and routine jobs. pricing changes fast. check openrouter ai for current rates before committing. Which is your favourite model and for what task? full 15 levels breakdown in the article 👇

YanXbt

17,138 просмотров • 3 месяцев назад

China is making Dario Amodei's AI slowdown proposal worthless. The thing doing it is a 510 GB file that anyone can download for free. Two days before that essay went out, DeepSeek shipped a model called V4.1 Flash. The weights went straight onto Hugging Face under an MIT license, which allows commercial use, modification and redistribution with no royalties owed to anybody. On its own model card, V4.1 Flash beats OpenAI's GPT-5.6 Sol and Anthropic's Claude Opus 5 on four of the five hardest agentic benchmarks: - DeepSWE v1.1: 74.2 against Sol's 73.0 - AutomationBench: 54.8 against 45.8 - Agent's Last Exam: 31.8 against 26.7 - CyberGym: 88.1 against 84.5 CyberGym is the cybersecurity one. So the exact capability Amodei is asking the industry to pace is now a PUBLIC DOWNLOAD LINK. Now look at the price... OpenAI listed GPT-6 Astra on September 3 at $10 per million input tokens and $50 per million output. DeepSeek listed V4.1 Flash seven days later at 15 cents and 60 cents off-peak. Cached input runs at a third of a cent. And the reasoning benchmarks tell the same story over a longer window. Scoring 87.5% on ARC-AGI-1 cost roughly $4,560 per task in December 2024. 20 months later the same score cost about 30 cents. DeepSeek's earlier Flash build scores higher than that, 89.0%, for 2 cents. On OpenDesign's arena on September 9, DeepSeek landed 1.5 points behind OpenAI's newest model at $0.023 per task against $1.61. 1.5 points. At 70 times the price. That's the American "lead“, measured this month. And that number is what breaks the entire proposal. The essay caps the permitted slowdown at the size of the lead, because slowing by more than that lets Chinese projects pull ahead. So the slowdown Anthropic, OpenAI, Google and xAI are allowed to take is 1.5 points wide. V4.1 Flash carries 552 billion parameters but switches on only 8 billion of them for each word it reads, which is why it runs for almost nothing. It trained on 45 trillion tokens. It's sitting on Hugging Face right now. A file on a hard drive signs nothing. Nobody can UNDOWNLOAD it. And the distillation crackdown in Dario‘s plan also does nothing here either, because nothing was distilled. DeepSeek published the weights outright. There's no theft to prosecute and no copy to trace. They gave it away on purpose. DeepSeek's Pro model already hit 80.6% on SWE-bench Verified back in April, at roughly a thirty-fourth of the price of the American flagships. This has been happening all year. The frontier still wins the hardest work. On Terminal-Bench 4.0, OpenAI reports 57.9 for its newest model against the 31.2 DeepSeek reports for this one. Those four benchmark wins come from DeepSeek's own card, and nobody has replicated all of them under one shared protocol. If you look at it that way, the American labs still hold the dangerous end of the curve, and pacing that end is worth doing. But a speed limit only binds companies that can be sued in an American court. Everyone else just downloads. What exactly is Dario Amodei's slowdown protecting you from?

China is making Dario Amodei's AI slowdown proposal worthless. The thing doing it is a 510 GB file that anyone can download for free. Two days before that essay went out, DeepSeek shipped a model called V4.1 Flash. The weights went straight onto Hugging Face under an MIT license, which allows commercial use, modification and redistribution with no royalties owed to anybody. On its own model card, V4.1 Flash beats OpenAI's GPT-5.6 Sol and Anthropic's Claude Opus 5 on four of the five hardest agentic benchmarks: - DeepSWE v1.1: 74.2 against Sol's 73.0 - AutomationBench: 54.8 against 45.8 - Agent's Last Exam: 31.8 against 26.7 - CyberGym: 88.1 against 84.5 CyberGym is the cybersecurity one. So the exact capability Amodei is asking the industry to pace is now a PUBLIC DOWNLOAD LINK. Now look at the price... OpenAI listed GPT-6 Astra on September 3 at $10 per million input tokens and $50 per million output. DeepSeek listed V4.1 Flash seven days later at 15 cents and 60 cents off-peak. Cached input runs at a third of a cent. And the reasoning benchmarks tell the same story over a longer window. Scoring 87.5% on ARC-AGI-1 cost roughly $4,560 per task in December 2024. 20 months later the same score cost about 30 cents. DeepSeek's earlier Flash build scores higher than that, 89.0%, for 2 cents. On OpenDesign's arena on September 9, DeepSeek landed 1.5 points behind OpenAI's newest model at $0.023 per task against $1.61. 1.5 points. At 70 times the price. That's the American "lead“, measured this month. And that number is what breaks the entire proposal. The essay caps the permitted slowdown at the size of the lead, because slowing by more than that lets Chinese projects pull ahead. So the slowdown Anthropic, OpenAI, Google and xAI are allowed to take is 1.5 points wide. V4.1 Flash carries 552 billion parameters but switches on only 8 billion of them for each word it reads, which is why it runs for almost nothing. It trained on 45 trillion tokens. It's sitting on Hugging Face right now. A file on a hard drive signs nothing. Nobody can UNDOWNLOAD it. And the distillation crackdown in Dario‘s plan also does nothing here either, because nothing was distilled. DeepSeek published the weights outright. There's no theft to prosecute and no copy to trace. They gave it away on purpose. DeepSeek's Pro model already hit 80.6% on SWE-bench Verified back in April, at roughly a thirty-fourth of the price of the American flagships. This has been happening all year. The frontier still wins the hardest work. On Terminal-Bench 4.0, OpenAI reports 57.9 for its newest model against the 31.2 DeepSeek reports for this one. Those four benchmark wins come from DeepSeek's own card, and nobody has replicated all of them under one shared protocol. If you look at it that way, the American labs still hold the dangerous end of the curve, and pacing that end is worth doing. But a speed limit only binds companies that can be sued in an American court. Everyone else just downloads. What exactly is Dario Amodei's slowdown protecting you from?

Ricardo

402,410 просмотров • 13 дней назад

HERMES AGENT + STRIPE PAYMENTS + NVIDIA NEMOTRON. YOUR AGENT CAN NOW RUN A BUSINESS. ACCEPT PAYMENTS. PAY FOR SERVICES. PROVISION ITS OWN INFRASTRUCTURE. ALL INSIDE A SECURITY SANDBOX. two years ago the question was: can an AI agent run a business autonomously? the answer shipped this week. Hermes already handles workflows: cron jobs, sub-agents, kanban orchestration, multi-profile pipelines, scheduled research. what it couldn't do: spend money and prove it's safe. Stripe solved the first problem. Nvidia solved the second. WHAT AUTONOMOUS BUSINESS OPERATIONS LOOK LIKE: → customer sends a request via email → agent reads, scopes the project, estimates cost → provisions the infrastructure it needs (pays via Stripe, you approve on your phone) → builds and deploys the deliverable → sends the result to the customer → creates a payment link via Stripe (Stripe API integration, separate from Link CLI) → tops off its own API credits when balance drops → reports daily costs and progress to your Telegram → all within security policies you set once you set the rules. the agent runs the operation. you review revenue reports. not tasks. this is already happening. Dark Factory: autonomous software factory. send an idea before bed. wake up to a deployed URL. live entry in the Hermes Accelerated Business Hackathon. HOW STRIPE MAKES THE AGENT FINANCIALLY AUTONOMOUS: Stripe Link CLI gives your agent a scoped wallet. not your credit card. one-time-use virtual cards. → agent finds a product or service it needs → creates a spend request via Stripe Link → you get a notification on your phone (Link app) → you review: merchant, amount, context → one tap to approve or reject → agent receives a one-time virtual card → completes the purchase → card expires after single use your real card details never enter agent context. never printed in chat. never exposed to the merchant. Hermes cannot self-approve. you confirm every spend. install: hermes install skills/optional/payments/stripe-link-cli link-cli auth login what the agent can pay for: → API credits (Nous Portal, OpenRouter) → SaaS subscriptions it needs for operations → domain names, hosting, cloud credits → products from any online store currently US only. HOW NVIDIA MAKES THE AGENT SAFE TO TRUST: an agent with spending authority and no security boundaries is a liability. NemoClaw solves this. three layers: 1. OPENSHELL (sandbox) kernel-level isolation. controls network, filesystem, syscalls. default deny. you whitelist what's allowed. agent tries to reach a blocked domain = rejected. agent has no idea it's sandboxed. 2. NEMOTRON (private models) open-weight models on your own hardware. Nemotron 3 Super 120B MoE (48GB+ VRAM). Nemotron 3 Nano 4B (8GB VRAM, edge). fully private. no data leaves your machine. without GPU: inference routes to cloud via Privacy Router. 3. PRIVACY ROUTER (automatic split) decides per query: local or cloud. private data → local Nemotron. general web research → Claude, GPT, Gemini. automatic. per query. no manual routing. install: export NEMOCLAW_AGENT=hermes curl -fsSL https:// www.nvidia. com/nemoclaw.sh | bash requires Docker. NemoClaw is alpha software. APIs may change. test in non-production first. THE FULL PICTURE: before this stack: → agent could work but couldn't pay for anything → agent could pay but couldn't be trusted → agent could be trusted but couldn't operate 24/7 now: → Hermes runs the business logic (workflows, memory, skills, cron, sub-agents) → Stripe runs the financial layer (Link CLI for spending, Stripe API for receiving) → NemoClaw runs the trust layer (sandbox, policies, private routing) → VPS keeps everything always on → Telegram keeps you in the loop TYPES OF BUSINESSES THIS ENABLES: → autonomous software factory (customer request → build → deploy → payment link) → content agency (brief → research → draft → deliver → bill) → lead generation service (scrape → qualify → outreach → book calls) → SaaS monitoring and maintenance (detect issues → fix → deploy → report) → e-commerce operations (inventory → pricing → fulfillment → support) each one: Hermes profiles handle the workflows. Stripe handles the payments (in and out). NemoClaw handles the security. you handle the strategy. THE HACKATHON: Hermes Agent Accelerated Business Hackathon with Nvidia and Stripe. cash prizes + Stripe credits + Nvidia DGX Spark. ends June 30. the goal: build agents that earn, spend, and run real operations autonomously. link in the Nous Research Discord. full Hermes architecture deep-dive in the article 👇

YanXbt

37,709 просмотров • 3 месяцев назад

HERMES AGENT IS PULLING AHEAD OF OPENCLAW. 8 FEATURES THAT WILL MAKE YOU SWITCH. 1. HERMES GETS SMARTER EVERY RUN. Hermes updates its own skills after every completed task. what worked gets saved. what failed gets refined. the Curator runs in the background every 7 days. prunes unused skills. consolidates duplicates. archives stale procedures to .archive/ (recoverable). your skill library stays clean without manual work. agents with 20+ self-created skills finish similar tasks ~40% faster. 2. CHECKPOINTS BEFORE EVERY FILE CHANGE. before Hermes touches your files, it snapshots the working directory with a shadow git store. if anything breaks: /rollback # restore last checkpoint /rollback 3 # go back 3 checkpoints config: → max 20 snapshots per project → max 10MB per file, 500MB total store → auto-prune after 7 days → opt-in: set checkpoints.enabled: true you can also restore a single file without affecting the rest of the directory. 3. STABLE RELEASES. Hermes ships fewer updates. the ones it ships go through 500+ PRs per release with community testing before merge. v0.16.0 had 874 commits and 542 merged PRs. the update system runs syntax validation after every pull. if anything breaks, it auto-rolls back to the last working state. 4. 27+ MESSAGING PLATFORMS. Telegram, Discord, Slack, WhatsApp, Signal, iMessage, SMS, Email, Teams, Matrix, and 17 more. one gateway process covers all of them. 5. PROFILE ISOLATION. separate agents with their own model, memory, skills, cron jobs, and SOUL.md. run a researcher on GPT-5.5 and a coder on Fable 5 simultaneously on the same machine. 6. KANBAN + DISPATCHER. task orchestration with 60-second dispatch cycle, zombie detection, heartbeat tracking, retry budgets. assign tasks to agents from a visual board. 7. /GOAL WITH JUDGE MODEL. persistent objectives across turns. a judge evaluates after each turn: done or continue. runs for hours autonomously. 8. BUILT-IN MIGRATION FROM OPENCLAW. hermes claw migrate one command moves your OpenClaw setup to Hermes. supports --dry-run to preview before changing anything. also migrates legacy Clawdbot and Moldbot setups. OpenClaw still has the bigger ecosystem and more community integrations. both tools are open source. both are actively developed. this is about which architecture compounds better over months of use. full Hermes SOUL . MD guide 👇

YanXbt

19,573 просмотров • 3 месяцев назад

i still don't understand why everyone is not using hermes bot yet, these are the must hermes skill people should install in bot 3200+ peoples bookmarked and using it already, and using this exact agent setup i made $2,443 last week in crypto signals alone. 17 skills on one agent. $0 extra. running a research floor that would cost a fund $300K a year in analyst salaries. the morning scan runs at 5am. i am not awake for it. 1. MONITOR scrapes real-time on-chain data, dev GitHubs, and unindexed Telegram alpha before CT finds it 2. RISK audits contract functions, mint rights, and LP locks, flags honeypots before entry 3. SNIPER places limit orders the exact second risk clearance passes 4. WHALE tracks smart money wallets and flags insider accumulation in real time 5. RUG monitors dev wallet activity 24/7 and exits the full position if LP is touched 6. EXIT trails stops dynamically and scales out as liquidity builds 7. SENTIMENT tracks social volume, momentum velocity, and influencer call timing 8. HEAD OF DESK never trades, routes data between every agent and brings me the one decision that needs a human 89 tokens scanned, 14 qualified setups, 5 executed trades while i was asleep. net result: +$8,400 after fees and zero bad fills. every skill has its own memory, browser, and local context in the cloud. the floor stays active with my laptop shut. the setup is dumber than it looks: > install Hermes and create your Head of Desk Bot > give the remaining 7 agents job descriptions like you're briefing new hires > run the workflow once on your screen while they watch > connect Telegram and wallet webhooks > no VPS, no code, no waiting on developers a trading floor used to mean 16-hour screen time, paid alpha groups, and constant fatigue. mine took one evening to set up. save this before your next trade. save the GUIDE

i still don't understand why everyone is not using hermes bot yet, these are the must hermes skill people should install in bot 3200+ peoples bookmarked and using it already, and using this exact agent setup i made $2,443 last week in crypto signals alone. 17 skills on one agent. $0 extra. running a research floor that would cost a fund $300K a year in analyst salaries. the morning scan runs at 5am. i am not awake for it. 1. MONITOR scrapes real-time on-chain data, dev GitHubs, and unindexed Telegram alpha before CT finds it 2. RISK audits contract functions, mint rights, and LP locks, flags honeypots before entry 3. SNIPER places limit orders the exact second risk clearance passes 4. WHALE tracks smart money wallets and flags insider accumulation in real time 5. RUG monitors dev wallet activity 24/7 and exits the full position if LP is touched 6. EXIT trails stops dynamically and scales out as liquidity builds 7. SENTIMENT tracks social volume, momentum velocity, and influencer call timing 8. HEAD OF DESK never trades, routes data between every agent and brings me the one decision that needs a human 89 tokens scanned, 14 qualified setups, 5 executed trades while i was asleep. net result: +$8,400 after fees and zero bad fills. every skill has its own memory, browser, and local context in the cloud. the floor stays active with my laptop shut. the setup is dumber than it looks: > install Hermes and create your Head of Desk Bot > give the remaining 7 agents job descriptions like you're briefing new hires > run the workflow once on your screen while they watch > connect Telegram and wallet webhooks > no VPS, no code, no waiting on developers a trading floor used to mean 16-hour screen time, paid alpha groups, and constant fatigue. mine took one evening to set up. save this before your next trade. save the GUIDE

Farea

56,047 просмотров • 1 месяц назад

qwen 3.8 max vs deepseek v4 flash 0731 vs kimi k3 vs gpt 5.6 sol – on rubik's cube and chess four frontier models built a rubik's cube stand and solved it, then built a chess board and played claude opus 5 on it the setup: Nous Research's hermes agent cli on OpenRouter tasks: 1. cube – build a 3d rubik's cube with a cli and a Three.js viewer, then solve an identical scrambled position on your own stand 2. chess – build a 3d chess stand, then play white against claude opus 5 as black, live, one move at a time. no engine, no solver, no opening book on either side. stockfish depth 14 grades every chess ply afterwards; neither player sees the score models: DeepSeek v4 flash 0731, OpenAI gpt-5.6 sol, Kimi.ai kimi k3, Qwen qwen 3.8 max gpt-5.6 sol and deepseek v4 flash solved their cubes – sol in 24 moves and seventeen seconds, deepseek in 32. qwen and kimi never got there, giving up at 96 and 207 moves then all four built chess stands and played white against claude opus 5 on them, and all four resigned: deepseek on move 13, sol on 19, kimi on 21, qwen holding out longest at 29 - build time, both stands #1 gpt-5.6 sol – 16m 43s #2 deepseek v4 flash – 97m 39s #3 kimi k3 – 166m 09s #4 qwen 3.8 max – 215m 08s - build attempts before a working stand #1 gpt-5.6 sol – 3 #2 qwen 3.8 max – 4 #3 kimi k3 – 4 #4 deepseek v4 flash – 5 - total tokens #1 gpt-5.6 sol – 6,713,754 #2 qwen 3.8 max – 17,272,507 #3 kimi k3 – 22,427,504 #4 deepseek v4 flash – 27,417,442 - total price #1 deepseek v4 flash – $0.557 #2 gpt-5.6 sol – $6.319 #3 qwen 3.8 max – $10.270 #4 kimi k3 – $16.667 observations: • deepseek v4 flash is the cheapest model here by a margin nobody else is near, and it got there while being the least efficient of the four. it burned 27.4m tokens – more than anyone, 5m more than kimi – and still finished both benchmarks for $0.557. that is $0.02 per million tokens against kimi's $0.74. it also needed the most passes to produce working stands, five, and that did not matter: all five deepseek passes together cost a thirtieth of kimi's two • so what deepseek cannot do is get it right the first time. what it can do is get it right the fifth time, for half a dollar. that is a different thing to be buying – not a good first draft, but the option to keep asking • gpt-5.6 sol is the opposite profile and the strongest of the four on pure efficiency. 16m 43s to build both stands, 6.7m tokens, three passes – under 40% of the next lowest token count and a quarter of deepseek's, on an eighth of qwen's clock. it also solved the cube fastest of anyone, 24 moves in seventeen seconds. sol is what you reach for when you want the answer now and can absorb $0.94 per million • sol's weakness is in what it does not check. its chess viewer deleted the capturing piece instead of the captured one, so pieces disappeared off the board mid-game – a defect the fifty-cent deepseek stand did not have. fast and terse turns out to be the same dial as fast and unverified • qwen 3.8 max is not the cheap open-weights option it gets treated as. $10.270 across the two benchmarks, second most expensive of the four, 18x deepseek, and by a distance the slowest – 215 minutes of build time, nearly thirteen times sol's. what the money buys is judgment: it played eighteen moves without a single error worth a hundredth of a pawn, then made exactly one bad move in the whole game, and averaged 44.6 centipawns lost across the longest game any of the four managed. it also could not solve a rubik's cube in 96 tries • kimi k3 is the one line with no reading that flatters it. most expensive at $16.667, last on the cube at 207 moves, last at chess at 478 centipawns lost per move. it is also the model that verified hardest – on the cube it wrote its own integrity check instead of trusting its output. that makes the result worse rather than better: the checking was real, and the reasoning underneath it still was not follow thehype. for 24/7 ai news, analysis and breakdowns

thehype.

84,777 просмотров • 1 месяц назад

7 tiny AI agent startup ideas you can start building today 1. The domain flipper agent. Monitors expired domain drops, scores them on backlinks and keywords, sends you a ranked list every morning. Buy for $10, flip for $3,000. I used to run this exact business manually with designers making logos for each domain. Now the whole thing is automated and 25x cheaper. 2. The local liquidation agent. Monitors restaurant closures and bankruptcy auctions in your city. Equipment worth $30k new sells for 10 cents on the dollar. Broker the deal for 15-30% fees with zero inventory risk. Works for dental, gym, and salon equipment too. 2. The hiring signal agent. Job postings are buying signals. Agent monitors boards daily, matches hiring patterns to what you sell, and sends draft outreach to your Slack every morning. Sell the leads to agencies or use them to build your own. 3. The sunset SaaS agent. Scans Product Hunt launches from 3-4 years ago that still have data and SEO traffic. Most founders have moved on and will be pumped to sell for cheap. Buy it. Rebuild the product as agent-first. 4. The dying app store agent. Finds apps that were top 100 three years ago, dropped to 500+, but still have 1,000+ reviews. Developer moved on. Product is validated. Acquire it. Relaunch with better monetization + AI features where it makes sense. 5. The competitive intel agent. Monitors 5 competitors while you sleep. Pricing changes, new pages, job postings, founder tweets. One-page brief by 7am. Sell as productized service or just use the intel. The idea framework behind all of these: 1. Think about any job where someone spends hours checking for updates or scanning listings 2. That's an agent 3. Build the agent that does the watching 4. You do the acting (or sell the watching to someone else) 5.Stack them. Each one is its own revenue stream. I built all of these using Genspark Claw in under 20 minutes each. Been testing it for last few weeks. I show you how to do it too in today's The Startup Ideas Podcast (SIP) 🧃 pod. Tiny agent ideas are interesting to me. Maybe you too. Some of the best businesses started off tiny. Watch

7 tiny AI agent startup ideas you can start building today 1. The domain flipper agent. Monitors expired domain drops, scores them on backlinks and keywords, sends you a ranked list every morning. Buy for $10, flip for $3,000. I used to run this exact business manually with designers making logos for each domain. Now the whole thing is automated and 25x cheaper. 2. The local liquidation agent. Monitors restaurant closures and bankruptcy auctions in your city. Equipment worth $30k new sells for 10 cents on the dollar. Broker the deal for 15-30% fees with zero inventory risk. Works for dental, gym, and salon equipment too. 2. The hiring signal agent. Job postings are buying signals. Agent monitors boards daily, matches hiring patterns to what you sell, and sends draft outreach to your Slack every morning. Sell the leads to agencies or use them to build your own. 3. The sunset SaaS agent. Scans Product Hunt launches from 3-4 years ago that still have data and SEO traffic. Most founders have moved on and will be pumped to sell for cheap. Buy it. Rebuild the product as agent-first. 4. The dying app store agent. Finds apps that were top 100 three years ago, dropped to 500+, but still have 1,000+ reviews. Developer moved on. Product is validated. Acquire it. Relaunch with better monetization + AI features where it makes sense. 5. The competitive intel agent. Monitors 5 competitors while you sleep. Pricing changes, new pages, job postings, founder tweets. One-page brief by 7am. Sell as productized service or just use the intel. The idea framework behind all of these: 1. Think about any job where someone spends hours checking for updates or scanning listings 2. That's an agent 3. Build the agent that does the watching 4. You do the acting (or sell the watching to someone else) 5.Stack them. Each one is its own revenue stream. I built all of these using Genspark Claw in under 20 minutes each. Been testing it for last few weeks. I show you how to do it too in today's The Startup Ideas Podcast (SIP) 🧃 pod. Tiny agent ideas are interesting to me. Maybe you too. Some of the best businesses started off tiny. Watch

GREG ISENBERG

78,487 просмотров • 4 месяцев назад

We release Needle 3: A Sliceable 8-29MB automation foundation model that can match DeepSeek V4 Flash. One set of weights, every depth from 2 to 20 layers a model of its own, 25-121M parameters at CQ2-bit, built on our Simple Attention Networks and running locally at up to 4k tokens/sec decode speed on a Raspberry Pi 5. Needle does not chat. Every turn is a function call: give it the tools your app exposes and it picks the right ones and fills every argument from what the user said, or hand it a schema and it returns a typed record. Ask for something no tool covers and you get an empty list, not a guess. That trade is lets 121M parameters trained on 360B tokens of structured data beat models 10x their size on mobile tool calls and match 2-3x bigger models on structured JSON extraction. It runs on mobiles, wearables, smart home devices, small robots and microcontrollers, with prebuilt engines for macOS, Linux, Windows, Android, iOS, watchOS, tvOS, the browser and WASI hosts. Try it in your browser:

We release Needle 3: A Sliceable 8-29MB automation foundation model that can match DeepSeek V4 Flash. One set of weights, every depth from 2 to 20 layers a model of its own, 25-121M parameters at CQ2-bit, built on our Simple Attention Networks and running locally at up to 4k tokens/sec decode speed on a Raspberry Pi 5. Needle does not chat. Every turn is a function call: give it the tools your app exposes and it picks the right ones and fills every argument from what the user said, or hand it a schema and it returns a typed record. Ask for something no tool covers and you get an empty list, not a guess. That trade is lets 121M parameters trained on 360B tokens of structured data beat models 10x their size on mobile tool calls and match 2-3x bigger models on structured JSON extraction. It runs on mobiles, wearables, smart home devices, small robots and microcontrollers, with prebuilt engines for macOS, Linux, Windows, Android, iOS, watchOS, tvOS, the browser and WASI hosts. Try it in your browser:

Cactus Compute

404,075 просмотров • 9 дней назад

1/5 I tested MiniMax H3 image-to-video locally on an RTX 4060 Ti 16GB using the smaller pruned INT8 model. 30 vertical videos generated. After removing the weaker results, I kept 23 strong candidates. The surprising part: 8 steps looked genuinely good. 2/5 Test settings: • 576×1024 vertical • 5-second videos • Native 24 FPS • Euler + simple scheduler • Sage Attention enabled • 8 steps, EasyCache disabled Average generation time was about 3 minutes per video. 3/5 Compared with my 20-step + EasyCache test: • 20 steps: 242 sec/video • 8 steps: 181 sec/video • About 25% faster overall The quality penalty depended on the scene. Complex patterns sometimes showed artifacts, but most of the 23 selected videos looked remarkably good. Example concept: a massive supercell forming a waterspout as jets pass overhead. 4/5 Audio was the weak point. At 8 steps, many clips produced a similar muddy disaster-rumble, even though the audio files weren’t literal duplicates. The lower step count hurt audio denoising far more than it hurt the visuals. Repeating cues like wind, alarms and thunder also made the tracks feel too similar. 5/5 Hardware survived the test without evidence of damage: • Peak GPU temperature: 85°C • Peak power: 166W • GPU and VRAM ran near their limits • No WHEA errors or driver resets • Thermal/power limiting behaved normally Takeaway: 8-step H3 is surprisingly viable for local video—but native audio needs more steps or a separate audio workflow.

1/5 I tested MiniMax H3 image-to-video locally on an RTX 4060 Ti 16GB using the smaller pruned INT8 model. 30 vertical videos generated. After removing the weaker results, I kept 23 strong candidates. The surprising part: 8 steps looked genuinely good. 2/5 Test settings: • 576×1024 vertical • 5-second videos • Native 24 FPS • Euler + simple scheduler • Sage Attention enabled • 8 steps, EasyCache disabled Average generation time was about 3 minutes per video. 3/5 Compared with my 20-step + EasyCache test: • 20 steps: 242 sec/video • 8 steps: 181 sec/video • About 25% faster overall The quality penalty depended on the scene. Complex patterns sometimes showed artifacts, but most of the 23 selected videos looked remarkably good. Example concept: a massive supercell forming a waterspout as jets pass overhead. 4/5 Audio was the weak point. At 8 steps, many clips produced a similar muddy disaster-rumble, even though the audio files weren’t literal duplicates. The lower step count hurt audio denoising far more than it hurt the visuals. Repeating cues like wind, alarms and thunder also made the tracks feel too similar. 5/5 Hardware survived the test without evidence of damage: • Peak GPU temperature: 85°C • Peak power: 166W • GPU and VRAM ran near their limits • No WHEA errors or driver resets • Thermal/power limiting behaved normally Takeaway: 8-step H3 is surprisingly viable for local video—but native audio needs more steps or a separate audio workflow.

Spectro

16,505 просмотров • 1 месяц назад