Video yükleniyor...

Video Yüklenemedi

Ana Sayfaya Dön

MiniCPM5-2B beats Qwen3.5-4B in 34% less time 👾 We gave three models the same real-world tasks and compared their attempts using Q4_K_M quants Tests: • Build a playlist within strict timing rules • Trace a checkout failure through server logs • Check stock and draft a replacement with tool...

atomic_chat_hq's profile picture

atomic.chat

15,985 subscribers

21,657 görüntüleme • 10 gün önce •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

6 Yorum

atomic.chat profil fotoğrafı
atomic.chat10 gün önce

Run AI models locally ->

TolkienWindow profil fotoğrafı
TolkienWindow10 gün önce

they provided a lot of the training data, fine-tune info, and recipes, will you folks be taking advantage of those at all? :+)

AshutoshShrivastava profil fotoğrafı
AshutoshShrivastava10 gün önce

No way..

Colbert profil fotoğrafı
Colbert10 gün önce

Efficiency gains like that are huge for real-world agent workflows. Honestly, seeing smaller models punch that far above their weight class is wild.

Branding Waves profil fotoğrafı
Branding Waves10 gün önce

Real world task performance is far more useful than benchmark scores alone.

Artiment Index profil fotoğrafı
Artiment Index10 gün önce

Same Q4 tasks: MiniCPM5-2B vs Qwen3.5-4B, and they clock 34% less time on the smaller seat. Local-agent density you can reproduce — that’s the practitioner tell.

Benzer Videolar

Parameter scaling just broke in front of me. I gave 90 math problems as images to 10 local vision models, 3 runs each, only consistent answers across all 3 runs counted. Two findings: > Gemma 4 was the most consistent family, 31B took the crown at 89.6%. > But Qwen 3.5 4B came 2 answers behind it. A 4B. Ranked 2nd out of 10. Vision-math isn't one skill, it's two: reading the image, then solving it. The real lesson for anyone running local: small ≠ weak. If you're building agentic workflows, finding the right model for each task matters more than reaching for the biggest one. In this test, the 4B ran far faster because of its size, scored higher, and left VRAM free for the rest of your stack. Full results: 🥇 Gemma 4 31B — 242/270 (89.6%) 🥈 Qwen 3.5 4B — 240/270 (88.9%) 🥉 Gemma 4 E4B — 222/270 (82.2%) 🥉 Qwen 3.6 27B — 222/270 (82.2%) 5. Gemma 4 26B A4B — 216/270 (80.0%) 6. Qwen 3.5 2B — 201/270 (74.4%) 7. Gemma 4 E2B — 192/270 (71.1%) 8. Qwen 3.6 35B A3B — 192/270 (71.1%) 9. Qwen 3.5 9B — 168/270 (62.2%) 10. Qwen 3.5 0.8B — 45/270 (16.7%) All GGUF + mmproj, Unsloth's Q6_K_XL quant.
0:44

Parameter scaling just broke in front of me. I gave 90 math problems as images to 10 local vision models, 3 runs each, only consistent answers across all 3 runs counted. Two findings: > Gemma 4 was the most consistent family, 31B took the crown at 89.6%. > But Qwen 3.5 4B came 2 answers behind it. A 4B. Ranked 2nd out of 10. Vision-math isn't one skill, it's two: reading the image, then solving it. The real lesson for anyone running local: small ≠ weak. If you're building agentic workflows, finding the right model for each task matters more than reaching for the biggest one. In this test, the 4B ran far faster because of its size, scored higher, and left VRAM free for the rest of your stack. Full results: 🥇 Gemma 4 31B — 242/270 (89.6%) 🥈 Qwen 3.5 4B — 240/270 (88.9%) 🥉 Gemma 4 E4B — 222/270 (82.2%) 🥉 Qwen 3.6 27B — 222/270 (82.2%) 5. Gemma 4 26B A4B — 216/270 (80.0%) 6. Qwen 3.5 2B — 201/270 (74.4%) 7. Gemma 4 E2B — 192/270 (71.1%) 8. Qwen 3.6 35B A3B — 192/270 (71.1%) 9. Qwen 3.5 9B — 168/270 (62.2%) 10. Qwen 3.5 0.8B — 45/270 (16.7%) All GGUF + mmproj, Unsloth's Q6_K_XL quant.

stevibe

22,882 görüntüleme • 4 ay önce

New Muse Glimmer 30B destroyed Gemma 4 31B at making retro arcade games! We gave three models the same task and compared one-shot outputs. Muse Glimmer 30B, Gemma 4 31B and Qwen3.6 27B each ran locally on its own RTX 5090 with 32GB VRAM Tasks: - Space Invaders - Tetris - Arkanoid with multiball Outputs: Muse Glimmer 30B: 83.4K tokens, 17.7 min Qwen3.6 27B: 23.7K tokens, 5.4 min Gemma 4 31B: 17.8K tokens, 4.5 min Glimmer's games play like the originals. The alien fleet speeds up with every kill, the blocks lock and lines clear, the bricks break. Nothing froze or crashed. Gemma broke on Tetris, its bot piles every piece against the left wall until the stack hits the ceiling with half the well empty. Qwen finished all three but some of its Arkanoid balls phase straight through the paddle. Run Muse Glimmer yourself in Atomic Chat from day 0!
0:26

New Muse Glimmer 30B destroyed Gemma 4 31B at making retro arcade games! We gave three models the same task and compared one-shot outputs. Muse Glimmer 30B, Gemma 4 31B and Qwen3.6 27B each ran locally on its own RTX 5090 with 32GB VRAM Tasks: - Space Invaders - Tetris - Arkanoid with multiball Outputs: Muse Glimmer 30B: 83.4K tokens, 17.7 min Qwen3.6 27B: 23.7K tokens, 5.4 min Gemma 4 31B: 17.8K tokens, 4.5 min Glimmer's games play like the originals. The alien fleet speeds up with every kill, the blocks lock and lines clear, the bricks break. Nothing froze or crashed. Gemma broke on Tetris, its bot piles every piece against the left wall until the stack hits the ceiling with half the well empty. Qwen finished all three but some of its Arkanoid balls phase straight through the paddle. Run Muse Glimmer yourself in Atomic Chat from day 0!

atomic.chat

39,067 görüntüleme • 1 ay önce

You need to know 4 things to run local AI on your own machine: 1) The model is the brain file. Gemma, Llama, Mistral, Qwen, Phi. 2) The warehouse is where you find the model. Hugging Face. 3) The software is what runs the model. LM Studio for the first try, Ollama when you build apps. 4) The workflow is the product you make from the other three. The size of the model decides the hardware: - 2B to 4B for phones and small tasks - 12B for the middle ground - 26B to 31B for a workstation Quantization compresses the model until it fits on a normal machine: - Q4 is easier to run - Q8 keeps more quality, but it needs more memory Q4 costs you a little quality but it saves you a $5,000 workstation. Start on the spare laptop from 2021. Open Hugging Face and read one model card slowly.
7:20

You need to know 4 things to run local AI on your own machine: 1) The model is the brain file. Gemma, Llama, Mistral, Qwen, Phi. 2) The warehouse is where you find the model. Hugging Face. 3) The software is what runs the model. LM Studio for the first try, Ollama when you build apps. 4) The workflow is the product you make from the other three. The size of the model decides the hardware: - 2B to 4B for phones and small tasks - 12B for the middle ground - 26B to 31B for a workstation Quantization compresses the model until it fits on a normal machine: - Q4 is easier to run - Q8 keeps more quality, but it needs more memory Q4 costs you a little quality but it saves you a $5,000 workstation. Start on the spare laptop from 2021. Open Hugging Face and read one model card slowly.

The Startup Ideas Podcast (SIP) 🧃

10,547 görüntüleme • 11 gün önce

🚨 GMI estimated ~$0.11 in savings on one 3-step Python coding workflow. I gave GMI Router the same e-commerce codebase and split the work into 3 requests: 👉 Code analysis → Cost Mode → DeepSeek-V4-Flash → ~$0.046 saved 👉 Debugging → Balanced Mode → DeepSeek-V4-Flash → ~$0.042 saved 👉 Implementation + testing → Quality Mode → DeepSeek-V4-Flash → ~$0.019 saved Total: ~$0.11 estimated savings vs Claude Opus 4.8, based on GMI's estimated token usage. But here's what caught my attention: Router selected the same model for all 3 tasks — even though I changed the mode each time. And that's actually the point. GMI Cloud 's Router is cache-aware. When the same model can handle the next request, sticking with it can help preserve cache reuse and reduce cost. There's another reason: These tasks weren't especially difficult. So even in Quality Mode, Router didn't need to move to a heavier model. The result: 1 codebase. 3 tasks. 3 routing modes. 1 lightweight model. ~$0.11 estimated savings. This is why request-level routing is more interesting than simply assigning a different model to every task.
1:11

🚨 GMI estimated ~$0.11 in savings on one 3-step Python coding workflow. I gave GMI Router the same e-commerce codebase and split the work into 3 requests: 👉 Code analysis → Cost Mode → DeepSeek-V4-Flash → ~$0.046 saved 👉 Debugging → Balanced Mode → DeepSeek-V4-Flash → ~$0.042 saved 👉 Implementation + testing → Quality Mode → DeepSeek-V4-Flash → ~$0.019 saved Total: ~$0.11 estimated savings vs Claude Opus 4.8, based on GMI's estimated token usage. But here's what caught my attention: Router selected the same model for all 3 tasks — even though I changed the mode each time. And that's actually the point. GMI Cloud 's Router is cache-aware. When the same model can handle the next request, sticking with it can help preserve cache reuse and reduce cost. There's another reason: These tasks weren't especially difficult. So even in Quality Mode, Router didn't need to move to a heavier model. The result: 1 codebase. 3 tasks. 3 routing modes. 1 lightweight model. ~$0.11 estimated savings. This is why request-level routing is more interesting than simply assigning a different model to every task.

Anthara Fairooz

28,520 görüntüleme • 23 gün önce

DFlash makes Qwen 2.2x faster with no quality loss! We ran the same Qwen3.6-27B locally three ways on one RTX 6000: baseline, MTP, DFlash. The tasks only differ in one thing - how predictable the next word is: quicksort, describe a file in JSON, a logic puzzle, a sci-fi story. Outputs: Baseline: 44 tok/s · 1.00x MTP: 65 tok/s · 1.45x · 71% accepted DFlash: 98 tok/s · 2.20x · 30% accepted Baseline writes one token per step. MTP works inside the model itself and guesses 3 tokens ahead. DFlash is a separate small model that writes 15 tokens at once, and the big model only checks them. In JSON the same words repeat all the time, so most guesses were right: 152 tok/s, 3.4x speedup. In the story 9 guesses out of 10 were wrong. DFlash did all that extra work for nothing and became slower than baseline: 42 vs 44 tok/s. MTP guesses only 3 tokens, so a wrong guess costs very little: 46 tok/s and the win in that round. The output is identical in all three modes - DFlash is the pick for tasks with predictable output, like coding, and for chat and creative writing MTP works better. DFlash is now natively integrated into Atomic Chat on llama.cpp - speed up your Qwen models!
0:26

DFlash makes Qwen 2.2x faster with no quality loss! We ran the same Qwen3.6-27B locally three ways on one RTX 6000: baseline, MTP, DFlash. The tasks only differ in one thing - how predictable the next word is: quicksort, describe a file in JSON, a logic puzzle, a sci-fi story. Outputs: Baseline: 44 tok/s · 1.00x MTP: 65 tok/s · 1.45x · 71% accepted DFlash: 98 tok/s · 2.20x · 30% accepted Baseline writes one token per step. MTP works inside the model itself and guesses 3 tokens ahead. DFlash is a separate small model that writes 15 tokens at once, and the big model only checks them. In JSON the same words repeat all the time, so most guesses were right: 152 tok/s, 3.4x speedup. In the story 9 guesses out of 10 were wrong. DFlash did all that extra work for nothing and became slower than baseline: 42 vs 44 tok/s. MTP guesses only 3 tokens, so a wrong guess costs very little: 46 tok/s and the win in that round. The output is identical in all three modes - DFlash is the pick for tasks with predictable output, like coding, and for chat and creative writing MTP works better. DFlash is now natively integrated into Atomic Chat on llama.cpp - speed up your Qwen models!

atomic.chat

83,110 görüntüleme • 2 ay önce

Excited to introduce SOLE-R1, a video-language reasoning model for zero-shot reward prediction for robot manipulation tasks! SOLE-R1 reasoning can serve as the SOLE signal for learning new tasks (completely from scratch) through online RL - i.e., robots start with random actions and learn previously unseen tasks guided only by SOLE-R1 rewards, without any demonstrations, ground-truth rewards, success indicators, or task-specific tuning. SOLE-R1 significantly outperforms strong baselines (e.g., Robometer, RoboReward, TOPReward, GPT-5, Gemini-3-Pro) in zero-shot online RL when evaluated across 40 tasks - including a real-world tabletop manipulation setting and 4 sim environments (LIBERO, ManiSkill, Meta-World, RoboSuite). We open source all models, training data, and code. Website, demos, and paper at: 🧵 (1/6)
1:15

Excited to introduce SOLE-R1, a video-language reasoning model for zero-shot reward prediction for robot manipulation tasks! SOLE-R1 reasoning can serve as the SOLE signal for learning new tasks (completely from scratch) through online RL - i.e., robots start with random actions and learn previously unseen tasks guided only by SOLE-R1 rewards, without any demonstrations, ground-truth rewards, success indicators, or task-specific tuning. SOLE-R1 significantly outperforms strong baselines (e.g., Robometer, RoboReward, TOPReward, GPT-5, Gemini-3-Pro) in zero-shot online RL when evaluated across 40 tasks - including a real-world tabletop manipulation setting and 4 sim environments (LIBERO, ManiSkill, Meta-World, RoboSuite). We open source all models, training data, and code. Website, demos, and paper at: 🧵 (1/6)

Philip Schroeder

19,275 görüntüleme • 3 ay önce

Atomic Agent beat Hermes on GAIA: 69.8% vs 58.5%, and it was 1.6x faster! We ran both agents through the full GAIA Level 1 benchmark, 53 real-world tasks, same 4-bit qwen-3.6-35b on the same Apple M4 Max. Results: ✦ Atomic Agent: 37 of 53 solved, done in 3h 12m ✦ Hermes Agent: 31 of 53 solved, took 5h 10m Atomic solved 6 more tasks and finished nearly 2 hours sooner. Hermes ran into the 900s timeout on 7 tasks; Atomic on just 2. Hermes burned 71% of its total time on tasks it still failed, Atomic, 48%. Where it showed: ✦ Audre Lorde poem, which stanza is indented: Atomic pushed through a dead source, switched tools, and answered in 7.6 min. Hermes ran the full clock and returned a blank. ✦ Vietnamese specimens, which city they ended up in: Atomic pulled it from the first source and normalized the answer in 33s. Hermes spent 7.3 min and never answered. ✦ The dinosaur featured-article nominator: Atomic walked the Wikipedia chain to "FunkMonk" in 57s. Hermes guessed a wrong name after 11 min. Atomic keeps a byte-stable prompt prefix, so llama-server reuses the KV-cache instead of re-encoding the whole context every turn, and it emits one JSON array of tool calls per inference, then compresses results back instead of pasting them in full, so the context never balloons and a small model stays sharp deep into a task. On top of that a no-progress guard vetoes repeated identical tool calls (warn at 3, hard veto at 5) and forces a reply, so Atomic never sinks 15 minutes into re-scanning one page the way Hermes did. Both agents missed some of the same questions, and on a few Hermes got there and Atomic did not, usually format slips where Atomic computed the right number but printed the working instead of the bare value. But on identical hardware and identical weights, the runtime that reuses its cache and refuses to spin came out ahead on accuracy and speed. Getting this from the runtime alone is wild. Run the same 53 GAIA tasks on Atomic Agent!
0:27

Atomic Agent beat Hermes on GAIA: 69.8% vs 58.5%, and it was 1.6x faster! We ran both agents through the full GAIA Level 1 benchmark, 53 real-world tasks, same 4-bit qwen-3.6-35b on the same Apple M4 Max. Results: ✦ Atomic Agent: 37 of 53 solved, done in 3h 12m ✦ Hermes Agent: 31 of 53 solved, took 5h 10m Atomic solved 6 more tasks and finished nearly 2 hours sooner. Hermes ran into the 900s timeout on 7 tasks; Atomic on just 2. Hermes burned 71% of its total time on tasks it still failed, Atomic, 48%. Where it showed: ✦ Audre Lorde poem, which stanza is indented: Atomic pushed through a dead source, switched tools, and answered in 7.6 min. Hermes ran the full clock and returned a blank. ✦ Vietnamese specimens, which city they ended up in: Atomic pulled it from the first source and normalized the answer in 33s. Hermes spent 7.3 min and never answered. ✦ The dinosaur featured-article nominator: Atomic walked the Wikipedia chain to "FunkMonk" in 57s. Hermes guessed a wrong name after 11 min. Atomic keeps a byte-stable prompt prefix, so llama-server reuses the KV-cache instead of re-encoding the whole context every turn, and it emits one JSON array of tool calls per inference, then compresses results back instead of pasting them in full, so the context never balloons and a small model stays sharp deep into a task. On top of that a no-progress guard vetoes repeated identical tool calls (warn at 3, hard veto at 5) and forces a reply, so Atomic never sinks 15 minutes into re-scanning one page the way Hermes did. Both agents missed some of the same questions, and on a few Hermes got there and Atomic did not, usually format slips where Atomic computed the right number but printed the working instead of the bare value. But on identical hardware and identical weights, the runtime that reuses its cache and refuses to spin came out ahead on accuracy and speed. Getting this from the runtime alone is wild. Run the same 53 GAIA tasks on Atomic Agent!

Atomic Agent

111,357 görüntüleme • 1 ay önce

Chinese researchers did it again! OpenBMB just open-sourced MiniCPM5-2B, a dense 2B-parameter model built for reasoning, coding, and tool use on resource-constrained hardware. Artificial Analysis ranked it highest among models under 4B in its Agentic Index comparison. It scored 20, while Granite 4.2 8B scored 9. The model is particularly strong at coding and tool calling, so I tested both capabilities locally. I pulled it onto my machine, connected it to a constrained CI repair agent, and gave it one issue: > A customer reports that retrying checkout with the same idempotency key returns a larger total. The first request returns $109, while the retry returns $118. Find the root cause, fix it without changing the public API contract, and verify the complete test suite. The Python checkout service had 18 tests. Sixteen passed, while two failed on the retry path. The agent could list files, search code, read selected ranges, run approved tests, apply a patch, and inspect its diff. It reproduced the failure, then followed the checkout and idempotency paths through the repository. The model found that shipping was added to mutable order state before the cached result was checked. On retry, the same order already contained shipping, so the calculation added it again. It generated a narrow patch that moved the idempotency check ahead of the mutation without changing the public API. The agent ran the targeted tests and the complete suite. All 18 tests passed. The model was never told which file contained the issue or what change to make. Each test result, search result, and code inspection determined its next action. The video below shows the full trajectory, including the investigation, tool calls, generated patch, diff, and final verification. Everything ran 100% locally on my machine throughout the run. MiniCPM5-2B supports llama.cpp, Ollama, vLLM, SGLang, iOS, Android, and HarmonyOS for local deployment. The model weights, training recipes, reasoning datasets, and UltraX data-refinement system are open-source. GitHub Repo: A 2B model can now inspect a repository, reason across multiple files, modify code, and verify its patch while remaining small enough to target local hardware.
0:29

Chinese researchers did it again! OpenBMB just open-sourced MiniCPM5-2B, a dense 2B-parameter model built for reasoning, coding, and tool use on resource-constrained hardware. Artificial Analysis ranked it highest among models under 4B in its Agentic Index comparison. It scored 20, while Granite 4.2 8B scored 9. The model is particularly strong at coding and tool calling, so I tested both capabilities locally. I pulled it onto my machine, connected it to a constrained CI repair agent, and gave it one issue: > A customer reports that retrying checkout with the same idempotency key returns a larger total. The first request returns $109, while the retry returns $118. Find the root cause, fix it without changing the public API contract, and verify the complete test suite. The Python checkout service had 18 tests. Sixteen passed, while two failed on the retry path. The agent could list files, search code, read selected ranges, run approved tests, apply a patch, and inspect its diff. It reproduced the failure, then followed the checkout and idempotency paths through the repository. The model found that shipping was added to mutable order state before the cached result was checked. On retry, the same order already contained shipping, so the calculation added it again. It generated a narrow patch that moved the idempotency check ahead of the mutation without changing the public API. The agent ran the targeted tests and the complete suite. All 18 tests passed. The model was never told which file contained the issue or what change to make. Each test result, search result, and code inspection determined its next action. The video below shows the full trajectory, including the investigation, tool calls, generated patch, diff, and final verification. Everything ran 100% locally on my machine throughout the run. MiniCPM5-2B supports llama.cpp, Ollama, vLLM, SGLang, iOS, Android, and HarmonyOS for local deployment. The model weights, training recipes, reasoning datasets, and UltraX data-refinement system are open-source. GitHub Repo: A 2B model can now inspect a repository, reason across multiple files, modify code, and verify its patch while remaining small enough to target local hardware.

Akshay 🚀

313,956 görüntüleme • 8 gün önce

Just gave a demo at Every 🪨's Vibe Code Camp showing how I use a kanban board to stay sane while overseeing Claude Code agents😎 Here's a quick recap video showing the whole flow: 1) Monitor progress on all tasks in one view 2) See when Claude is blocked, and respond in context 3) Read Claude's plans with nice formatting, not raw Markdown 4) Use AI to process voice notes into structured tasks 5) Get explanations of code changes... with an interactive quiz to check if you understood 🙃 I've had mixed feelings about kanban workflows in the past, but as the agents get stronger I'm warming up to them—especially for churning through small well-specified tasks. In that context, it feels like the challenge is managing my working memory for what's going on. And having a stable "zoomed-out map" of all the tasks really helps me with offloading some of that cognitive load to a visualization of everything that's going on!
7:47

Just gave a demo at Every 🪨's Vibe Code Camp showing how I use a kanban board to stay sane while overseeing Claude Code agents😎 Here's a quick recap video showing the whole flow: 1) Monitor progress on all tasks in one view 2) See when Claude is blocked, and respond in context 3) Read Claude's plans with nice formatting, not raw Markdown 4) Use AI to process voice notes into structured tasks 5) Get explanations of code changes... with an interactive quiz to check if you understood 🙃 I've had mixed feelings about kanban workflows in the past, but as the agents get stronger I'm warming up to them—especially for churning through small well-specified tasks. In that context, it feels like the challenge is managing my working memory for what's going on. And having a stable "zoomed-out map" of all the tasks really helps me with offloading some of that cognitive load to a visualization of everything that's going on!

Geoffrey Litt

78,427 görüntüleme • 8 ay önce

glm 5.3 vs qwen 3.8 vs gemini 3.7 vs deepseek v4 flash four models designed and built three structures each on a physics-backed site, with no dimensions anywhere in the brief the setup: our own agent loop on OpenRouter, a construction site as the tool set – footings, walls, arches, roofs, scaffold, a lamp. the site enforces physics and nothing else: unsupported brick falls, a roof needs walls under it, a worker reaches 3.2 m above whatever he stands on, an arch needs centring until the keystone is set, concrete cures before it carries. no budget ceiling – material cost is tallied and reported, never blocked. tasks: 1. house – a plot and a palette, no plan. shape, height and material are the model's call 2. lighthouse – a headland cut by a gully, with a rock stack standing 30 m offshore. the lamp must burn, it must be the highest thing built, and the keeper must be able to walk to it 3. bridge – a river with one islet and banks at different heights. cross it however you want models: Z.ai glm 5.3 flash, Qwen qwen 3.8 flash, Google DeepMind gemini 3.7 flash, DeepSeek v4 flash vision all twelve objects were finished and signed off by the models themselves. tallest lighthouse is qwen's at 38.4 m, planted on the offshore stack with a bridge run out to it – the only model that read the site that way. deepseek signed off its bridge on an empty riverbed: 0 bricks, 107 minutes, $1.16m of material tallied - total cost, three builds #1 glm 5.3 flash – $0.201 #2 gemini 3.7 flash – $0.871 #3 qwen 3.8 flash – $1.058 #4 deepseek v4 flash – $1.567 - wall clock, three builds #1 gemini 3.7 flash – 91m #2 glm 5.3 flash – 228m #3 deepseek v4 flash – 502m #4 qwen 3.8 flash – 912m - total tokens #1 gemini 3.7 flash – 3,567,052 #2 glm 5.3 flash – 4,732,748 #3 qwen 3.8 flash – 13,469,333 #4 deepseek v4 flash – 18,230,076 - defects logged by the site #1 deepseek v4 flash – 59 #2 gemini 3.7 flash – 132 #3 glm 5.3 flash – 221 #4 qwen 3.8 flash – 350 - material tallied across three builds #1 gemini 3.7 flash – $359,884 #2 glm 5.3 flash – $583,358 #3 deepseek v4 flash – $1,327,484 #4 qwen 3.8 flash – $2,188,625 observations: • glm is the cheap one and nothing here is close – $0.201 for three buildings, $0.042 per million tokens, 6x under gemini's rate • what glm spends it on is bulk, not care: 166,228 bricks in one house and 156 defect weight, the worst single object in the set • gemini is the efficiency line – 91 minutes and 3.57m tokens for all three and an eighth of qwen's clock • gemini also builds the smallest of everything. its lighthouse is 22.5 m against qwen's 38.4, its house 6.9 m against 19.3 • qwen is the maximalist: 1.18m bricks, $2.19m of material, tallest on all three tasks, and 912 minutes – 15 hours – to get there conclusion: twelve finished objects for $3.80 all in, and a 7.8x price spread between the cheapest model and the priciest! follow thehype. for 24/7 ai news, analysis and breakdowns
1:09

glm 5.3 vs qwen 3.8 vs gemini 3.7 vs deepseek v4 flash four models designed and built three structures each on a physics-backed site, with no dimensions anywhere in the brief the setup: our own agent loop on OpenRouter, a construction site as the tool set – footings, walls, arches, roofs, scaffold, a lamp. the site enforces physics and nothing else: unsupported brick falls, a roof needs walls under it, a worker reaches 3.2 m above whatever he stands on, an arch needs centring until the keystone is set, concrete cures before it carries. no budget ceiling – material cost is tallied and reported, never blocked. tasks: 1. house – a plot and a palette, no plan. shape, height and material are the model's call 2. lighthouse – a headland cut by a gully, with a rock stack standing 30 m offshore. the lamp must burn, it must be the highest thing built, and the keeper must be able to walk to it 3. bridge – a river with one islet and banks at different heights. cross it however you want models: Z.ai glm 5.3 flash, Qwen qwen 3.8 flash, Google DeepMind gemini 3.7 flash, DeepSeek v4 flash vision all twelve objects were finished and signed off by the models themselves. tallest lighthouse is qwen's at 38.4 m, planted on the offshore stack with a bridge run out to it – the only model that read the site that way. deepseek signed off its bridge on an empty riverbed: 0 bricks, 107 minutes, $1.16m of material tallied - total cost, three builds #1 glm 5.3 flash – $0.201 #2 gemini 3.7 flash – $0.871 #3 qwen 3.8 flash – $1.058 #4 deepseek v4 flash – $1.567 - wall clock, three builds #1 gemini 3.7 flash – 91m #2 glm 5.3 flash – 228m #3 deepseek v4 flash – 502m #4 qwen 3.8 flash – 912m - total tokens #1 gemini 3.7 flash – 3,567,052 #2 glm 5.3 flash – 4,732,748 #3 qwen 3.8 flash – 13,469,333 #4 deepseek v4 flash – 18,230,076 - defects logged by the site #1 deepseek v4 flash – 59 #2 gemini 3.7 flash – 132 #3 glm 5.3 flash – 221 #4 qwen 3.8 flash – 350 - material tallied across three builds #1 gemini 3.7 flash – $359,884 #2 glm 5.3 flash – $583,358 #3 deepseek v4 flash – $1,327,484 #4 qwen 3.8 flash – $2,188,625 observations: • glm is the cheap one and nothing here is close – $0.201 for three buildings, $0.042 per million tokens, 6x under gemini's rate • what glm spends it on is bulk, not care: 166,228 bricks in one house and 156 defect weight, the worst single object in the set • gemini is the efficiency line – 91 minutes and 3.57m tokens for all three and an eighth of qwen's clock • gemini also builds the smallest of everything. its lighthouse is 22.5 m against qwen's 38.4, its house 6.9 m against 19.3 • qwen is the maximalist: 1.18m bricks, $2.19m of material, tallest on all three tasks, and 912 minutes – 15 hours – to get there conclusion: twelve finished objects for $3.80 all in, and a 7.8x price spread between the cheapest model and the priciest! follow thehype. for 24/7 ai news, analysis and breakdowns

thehype.

26,360 görüntüleme • 22 gün önce

Can vision-language-action (VLA) models generalize to diverse OOD tasks and align with customized objectives? 🤔 🚀 We introduce GRAPE, a plug-and-play algorithm to generalize robot policies via preference alignment. GRAPE unfolds three benefits to boost the generalizability of VLAs: 👉1. GRAPE aligns VLAs on a trajectory level and endows the model with the ability for global decision-making, instead of merely cloning behavior; 👉2. GRAPE implicitly models reward from both successful and failed trials to boost generalizability to diverse tasks; 👉3. GRAPE adopts a scalable preference synthesis algorithm to rank trajectories with preferences that align with arbitrary objectives. Our experiments on a diverse array of real-world and simulated robotic tasks reveal: 1⃣GRAPE enhances the performance of state-of-the-art VLA models, increasing success rates on in-domain and unseen manipulation tasks by 51.79% and 60.36%; 2⃣GRAPE is versatile to be aligned with diverse objectives and reduce collision rates by 44.31% or rollout length by 11.15% when aligning towards safer or more efficient manipulation policy, respectively. Check out our full project for more details: 🔥 Paper: 🔥 Project: 🔥 Code:
1:06

Can vision-language-action (VLA) models generalize to diverse OOD tasks and align with customized objectives? 🤔 🚀 We introduce GRAPE, a plug-and-play algorithm to generalize robot policies via preference alignment. GRAPE unfolds three benefits to boost the generalizability of VLAs: 👉1. GRAPE aligns VLAs on a trajectory level and endows the model with the ability for global decision-making, instead of merely cloning behavior; 👉2. GRAPE implicitly models reward from both successful and failed trials to boost generalizability to diverse tasks; 👉3. GRAPE adopts a scalable preference synthesis algorithm to rank trajectories with preferences that align with arbitrary objectives. Our experiments on a diverse array of real-world and simulated robotic tasks reveal: 1⃣GRAPE enhances the performance of state-of-the-art VLA models, increasing success rates on in-domain and unseen manipulation tasks by 51.79% and 60.36%; 2⃣GRAPE is versatile to be aligned with diverse objectives and reduce collision rates by 44.31% or rollout length by 11.15% when aligning towards safer or more efficient manipulation policy, respectively. Check out our full project for more details: 🔥 Paper: 🔥 Project: 🔥 Code:

Huaxiu Yao

20,071 görüntüleme • 1 yıl önce

Diffusion Gemma is 4x faster, but makes 6x more mistakes! We benchmarked the new diffusion LLM against its autoregressive twin on a single H100 (FP8). We gave each the same three tasks: write a Steve Jobs biography, the history of Tetris, and the story of BeOS - every next topic less popular than the previous one. Then we fact-checked every claim in every answer. Gemma4 got 45 facts right, 5 wrong. DiffusionGemma got 33 right, 28 wrong. The less popular the topic, the worse it got: 4 mistakes on Jobs, 12 on Tetris, 12 on BeOS. It named Clara Clley as Steve Jobs' mother, invented a colleague for Pajitnov named Geri Gulovik and priced the BeBox at $9,999. The real one cost $1,600. Outputs: Gemma4 26B A4B: 218 tok/s · 15.1s total · 45 facts · 5 mistakes DiffusionGemma 26B A4B: 763 tok/s · 3.7s total · 33 facts · 28 mistakes The reason is simple. DiffusionGemma throws 256 tokens on the screen at once and polishes them pass after pass until the text sounds smooth. Smooth is all it cares about: a fake name, date or number sounds just as smooth as a real one, so it stays. Regular Gemma4 meanwhile writes one word at a time and checks every new word against everything before it. Google says it themselves in the launch post: quality is lower, use regular Gemma 4 when facts matter.
0:32

Diffusion Gemma is 4x faster, but makes 6x more mistakes! We benchmarked the new diffusion LLM against its autoregressive twin on a single H100 (FP8). We gave each the same three tasks: write a Steve Jobs biography, the history of Tetris, and the story of BeOS - every next topic less popular than the previous one. Then we fact-checked every claim in every answer. Gemma4 got 45 facts right, 5 wrong. DiffusionGemma got 33 right, 28 wrong. The less popular the topic, the worse it got: 4 mistakes on Jobs, 12 on Tetris, 12 on BeOS. It named Clara Clley as Steve Jobs' mother, invented a colleague for Pajitnov named Geri Gulovik and priced the BeBox at $9,999. The real one cost $1,600. Outputs: Gemma4 26B A4B: 218 tok/s · 15.1s total · 45 facts · 5 mistakes DiffusionGemma 26B A4B: 763 tok/s · 3.7s total · 33 facts · 28 mistakes The reason is simple. DiffusionGemma throws 256 tokens on the screen at once and polishes them pass after pass until the text sounds smooth. Smooth is all it cares about: a fake name, date or number sounds just as smooth as a real one, so it stays. Regular Gemma4 meanwhile writes one word at a time and checks every new word against everything before it. Google says it themselves in the launch post: quality is lower, use regular Gemma 4 when facts matter.

atomic.chat

77,289 görüntüleme • 3 ay önce

FDE is the hottest role in tech. Here's the 30-day plan to actually become one: Week 1: Build the loop - Agent looping, tool usage, guardrails - Context, memory, and an audit trail - Checkpoint: one working agent, one task. Week 2: Make it recover - JSON schemas, not freeform text - Failure modes for the 1,000 ways it breaks - "There's only one way something goes right. A thousand ways it goes wrong." Week 3: Make it measurable - Retry logic + a golden dataset for evals - Test cheaper models on sub-tasks - Score it on revenue, risk, cost Week 4: Defend it like an FDE - The pain points, the architecture, the economics - Then pitch it to real businesses Think: - Week 1 = build - Week 2 = survive - Week 3 = measure - Week 4 = sell That's the whole ladder.
6:57

FDE is the hottest role in tech. Here's the 30-day plan to actually become one: Week 1: Build the loop - Agent looping, tool usage, guardrails - Context, memory, and an audit trail - Checkpoint: one working agent, one task. Week 2: Make it recover - JSON schemas, not freeform text - Failure modes for the 1,000 ways it breaks - "There's only one way something goes right. A thousand ways it goes wrong." Week 3: Make it measurable - Retry logic + a golden dataset for evals - Test cheaper models on sub-tasks - Score it on revenue, risk, cost Week 4: Defend it like an FDE - The pain points, the architecture, the economics - Then pitch it to real businesses Think: - Week 1 = build - Week 2 = survive - Week 3 = measure - Week 4 = sell That's the whole ladder.

The Startup Ideas Podcast (SIP) 🧃

108,443 görüntüleme • 2 ay önce

ByteDance Seed delivered again. They released EdgeBench, to test whether AI agents can improve through experience, using 134 real-world tasks that run for at least 12 hours. The big deal is that it shifts AI evaluation from “what does the model already know?” to “can the model learn while doing real work?” Huge, because future AI agents will not just answer questions from training data. They will enter messy environments, use tools, make attempts, read feedback, fix mistakes, and slowly build better solutions. Most current benchmarks are too short for that, so they mostly test memory, coding skill, or one-shot reasoning. EdgeBench instead gives agents 12-hour real-world tasks with feedback loops, so it can measure whether the agent improves through experience. Each task has a local workspace for fast trial and error, plus a hidden judge that gives stronger feedback on submitted work, which is meant to feel closer to real expert work. The authors then ran frontier agents for about 38,000 total hours and tracked how their best score changed as they kept interacting with the task environment. The big result is that when scores are averaged across many tasks, learning follows a very clean log-sigmoid curve, meaning progress is slow, then faster, then starts to level off. They also found that newer agents seem to learn from environments much faster, with the top models roughly doubling their 2-hour learning speed every 3 months.
0:52

ByteDance Seed delivered again. They released EdgeBench, to test whether AI agents can improve through experience, using 134 real-world tasks that run for at least 12 hours. The big deal is that it shifts AI evaluation from “what does the model already know?” to “can the model learn while doing real work?” Huge, because future AI agents will not just answer questions from training data. They will enter messy environments, use tools, make attempts, read feedback, fix mistakes, and slowly build better solutions. Most current benchmarks are too short for that, so they mostly test memory, coding skill, or one-shot reasoning. EdgeBench instead gives agents 12-hour real-world tasks with feedback loops, so it can measure whether the agent improves through experience. Each task has a local workspace for fast trial and error, plus a hidden judge that gives stronger feedback on submitted work, which is meant to feel closer to real expert work. The authors then ran frontier agents for about 38,000 total hours and tracked how their best score changed as they kept interacting with the task environment. The big result is that when scores are averaged across many tasks, learning follows a very clean log-sigmoid curve, meaning progress is slow, then faster, then starts to level off. They also found that newer agents seem to learn from environments much faster, with the top models roughly doubling their 2-hour learning speed every 3 months.

Rohan Paul

14,309 görüntüleme • 2 ay önce

HERMES AGENT NOW RUNS A FULL BUSINESS FOR $8/MONTH. CONTENT. CODE. INBOX. ALL AGENTS. EVERYTHING WHILE YOU SLEEP. one user built this in one session. zero terminal commands. browser only. the team: → content agent: finds video topics, writes first drafts, helps with animations → code builder: writes website and SaaS code → ops agent: organizes and answers email total setup cost: $1. monthly runtime: $8 VPS + cheap model API. the trick is matching model to task. frontier models (Opus 4.8, GPT-5.5) for complex /goals only. $5/million tokens. mid-tier (Claude Sonnet, Haiku, Kimiko 2.6) for daily content and research work. budget (DeepSeek V4 Flash, MiniMax 2.7, Gemma 4, Qwen 3) for cron jobs and routine. run the initial setup on Opus. switch to DeepSeek for everything after. your bill drops 90% with zero quality loss on routine work. the agent team picks up tasks from kanban. each completed task creates a skill. week 3 the same workflows run faster than week 1. full breakdown on Hermes Os in the article 👇
14:53

HERMES AGENT NOW RUNS A FULL BUSINESS FOR $8/MONTH. CONTENT. CODE. INBOX. ALL AGENTS. EVERYTHING WHILE YOU SLEEP. one user built this in one session. zero terminal commands. browser only. the team: → content agent: finds video topics, writes first drafts, helps with animations → code builder: writes website and SaaS code → ops agent: organizes and answers email total setup cost: $1. monthly runtime: $8 VPS + cheap model API. the trick is matching model to task. frontier models (Opus 4.8, GPT-5.5) for complex /goals only. $5/million tokens. mid-tier (Claude Sonnet, Haiku, Kimiko 2.6) for daily content and research work. budget (DeepSeek V4 Flash, MiniMax 2.7, Gemma 4, Qwen 3) for cron jobs and routine. run the initial setup on Opus. switch to DeepSeek for everything after. your bill drops 90% with zero quality loss on routine work. the agent team picks up tasks from kanban. each completed task creates a skill. week 3 the same workflows run faster than week 1. full breakdown on Hermes Os in the article 👇

YanXbt

30,910 görüntüleme • 3 ay önce

A team tested Pi0, Pi0 Fast, Gr00t, and ACT on real robot arms in manufacturing tasks. (🔖 Bookmark this for later!) The task was precise: place thin rectangular frames from a messy stack into a holder. The team fine-tuned each model on 100 real trajectories and compared training time, inference speed, motion quality, and success rates. ⬇️ Here’s a breakdown of what they found Pi0 (Original) ✅ Strongest overall performance in precise pick-and-place ✅ High success rate even in edge cases ✅ Longest training time (~11 hours, ~$30 per run) ✅ Inference time of 80 ms causes short pauses between actions Despite delays, it handles complex scenarios well… solid for high-precision tasks, but slow to train. Gr00t ✅ Trains fast (~2 hours, ~$5 per run) ✅ Performs almost as well as Pi0 on large-object tasks ✅ Struggles with fine precision; random movement in some trials ✅ More training didn’t fix jitter or random offsets Best suited for tasks where exact precision isn’t critical. Not ready for manufacturing-grade accuracy without more tuning. Pi0 Fast ✅ Promised faster training, but results were underwhelming ✅ Training at 6 hours still showed low success rates ✅ Inference was slower than expected ✅ Not reliable for generalizing even slightly new tasks Currently too unstable for real-world deployment. Doesn’t live up to the “Fast” name yet. ACT (Baseline) ✅ 200MB model—lightweight, but limited ✅ Struggles with stacked objects or ambiguous scenes ✅ Success rates around 70% in best-case setups ✅ Can’t match newer models on precision or generalization Still a solid baseline, but clearly a generation behind in robustness. 🚨 Extra Notes All newer models share a common issue: •Inference takes longer than a frame (80 ms vs 33 ms), so robots “pause” between chunks. •This results in jittery movements, but not a dealbreaker unless tasks are time-sensitive. Language-conditioned tasks also fell short: after training on two labeled tasks, the model couldn’t generalize to a third unseen combination using only text prompts. ✅ The good news? These models adapt well to new robot arms with quick fine-tuning. ❌ The bad news? There’s still no plug-and-play solution for improving performance after deployment. Reinforcement learning or DAgger-style data collection during real-world operation may be the next big step, something many teams in robotics are actively working on.
0:51

A team tested Pi0, Pi0 Fast, Gr00t, and ACT on real robot arms in manufacturing tasks. (🔖 Bookmark this for later!) The task was precise: place thin rectangular frames from a messy stack into a holder. The team fine-tuned each model on 100 real trajectories and compared training time, inference speed, motion quality, and success rates. ⬇️ Here’s a breakdown of what they found Pi0 (Original) ✅ Strongest overall performance in precise pick-and-place ✅ High success rate even in edge cases ✅ Longest training time (~11 hours, ~$30 per run) ✅ Inference time of 80 ms causes short pauses between actions Despite delays, it handles complex scenarios well… solid for high-precision tasks, but slow to train. Gr00t ✅ Trains fast (~2 hours, ~$5 per run) ✅ Performs almost as well as Pi0 on large-object tasks ✅ Struggles with fine precision; random movement in some trials ✅ More training didn’t fix jitter or random offsets Best suited for tasks where exact precision isn’t critical. Not ready for manufacturing-grade accuracy without more tuning. Pi0 Fast ✅ Promised faster training, but results were underwhelming ✅ Training at 6 hours still showed low success rates ✅ Inference was slower than expected ✅ Not reliable for generalizing even slightly new tasks Currently too unstable for real-world deployment. Doesn’t live up to the “Fast” name yet. ACT (Baseline) ✅ 200MB model—lightweight, but limited ✅ Struggles with stacked objects or ambiguous scenes ✅ Success rates around 70% in best-case setups ✅ Can’t match newer models on precision or generalization Still a solid baseline, but clearly a generation behind in robustness. 🚨 Extra Notes All newer models share a common issue: •Inference takes longer than a frame (80 ms vs 33 ms), so robots “pause” between chunks. •This results in jittery movements, but not a dealbreaker unless tasks are time-sensitive. Language-conditioned tasks also fell short: after training on two labeled tasks, the model couldn’t generalize to a third unseen combination using only text prompts. ✅ The good news? These models adapt well to new robot arms with quick fine-tuning. ❌ The bad news? There’s still no plug-and-play solution for improving performance after deployment. Reinforcement learning or DAgger-style data collection during real-world operation may be the next big step, something many teams in robotics are actively working on.

Ilir Aliu

21,844 görüntüleme • 1 yıl önce

New Z.ai GLM-5.2 beats Kimi K2.7 Code on physics contest! We gave both models the same three prompts and asked them to build self contained HTML5 sims with real physics and no libraries: 1. Pool break 2. Block on a bed of springs 3. Galton board Outputs: GLM-5.2: 12,640 tokens Kimi K2.7 Code: 7,420 tokens GLM 5.2 nailed all three, and it did it with way more detail and polish. The break conserved momentum, the block bounced off the springs and the Galton beads spread into a clean bell curve. Kimi struggled on every scene: its block fell straight through the springs, its break didn't look realistic with the balls colliding all wrong, and on the Galton board its balls overlapped and piled into each other instead of spreading out
0:33

New Z.ai GLM-5.2 beats Kimi K2.7 Code on physics contest! We gave both models the same three prompts and asked them to build self contained HTML5 sims with real physics and no libraries: 1. Pool break 2. Block on a bed of springs 3. Galton board Outputs: GLM-5.2: 12,640 tokens Kimi K2.7 Code: 7,420 tokens GLM 5.2 nailed all three, and it did it with way more detail and polish. The break conserved momentum, the block bounced off the springs and the Galton beads spread into a clean bell curve. Kimi struggled on every scene: its block fell straight through the springs, its break didn't look realistic with the balls colliding all wrong, and on the Galton board its balls overlapped and piled into each other instead of spreading out

atomic.chat

60,899 görüntüleme • 3 ay önce

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
0:45

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 görüntüleme • 1 ay önce

New Fable 5 beats Opus 4.8 on real world physics simulations We gave both models the same three prompts and asked them to build self contained HTML5 sims with real physics and no libraries: 1. Chaotic double pendulum 2. Galton board 3. Water in a spinning drum (WCSPH) Generation cost Fable 5: $3.35 on 68.7k tokens, time 14m 47s Opus 4.8: $0.93 on 38.9k tokens, time 8m 10s Fable clearly did better on the water simulation, producing a much more solid and continuous body of water. Opus left larger gaps near the walls, scattered particles around the scene, and struggled to keep the fluid stable.
0:22

New Fable 5 beats Opus 4.8 on real world physics simulations We gave both models the same three prompts and asked them to build self contained HTML5 sims with real physics and no libraries: 1. Chaotic double pendulum 2. Galton board 3. Water in a spinning drum (WCSPH) Generation cost Fable 5: $3.35 on 68.7k tokens, time 14m 47s Opus 4.8: $0.93 on 38.9k tokens, time 8m 10s Fable clearly did better on the water simulation, producing a much more solid and continuous body of water. Opus left larger gaps near the walls, scattered particles around the scene, and struggled to keep the fluid stable.

atomic.chat

1,587,969 görüntüleme • 3 ay önce