Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

GLM-5.2 delivers a substantial leap in app development capabilities, which also represent demanding long-horizon tasks. Results: - GLM-5.1: 21/70 - GLM-5.2: 48/70 - Claude Fable 5: 56/70 That's more than a twofold improvement from GLM-5.1 to GLM-5.2. These come from an internal benchmark of 35 challenging mobile development tasks,... show more

Zixuan Li

18,267 subscribers

311,382 views • 25 days ago •via X (Twitter)

News & Politics Education Science & Technology

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

Introducing GLM 5.2 for autoresearch GLM 5.2 is the first open weights model we've tried on our autoresearch pipeline that's proven capable for real research tasks. With Fable 5's restrictions on research, having an open weights alternative is a huge win for open source Watch it carry out fully async vs colocated sync RL training on Harbor code contests across two 8xH100 nodes on top of SkyRL. Resolves setup issues, tracks runs to completion, and produces a full comparison of throughput and reward stability

Introducing GLM 5.2 for autoresearch GLM 5.2 is the first open weights model we've tried on our autoresearch pipeline that's proven capable for real research tasks. With Fable 5's restrictions on research, having an open weights alternative is a huge win for open source Watch it carry out fully async vs colocated sync RL training on Harbor code contests across two 8xH100 nodes on top of SkyRL. Resolves setup issues, tracks runs to completion, and produces a full comparison of throughput and reward stability

alphaXiv

268,617 views • 22 days ago

I tested GLM 5.2 on three real builds. A horror house game. A 3D stealth game. A Remotion marketing video. The results are exactly what I expect from Chinese open source coding models right now. The horror house has broken game logic. Collect all three keys, complete the objective, and the front door still won't open. The stealth game shipped with lighting so dark you can't see your own character. The Remotion video is rough. GLM 5.2 is faster than 5.1 and 10x cheaper than the frontier. But Opus 4.8 and GPT 5.5 are still in a different league for serious coding work. Chinese models keep missing the attention to detail that matters. I MISS FABLE 5 ALREADY. Full review live now.

I tested GLM 5.2 on three real builds. A horror house game. A 3D stealth game. A Remotion marketing video. The results are exactly what I expect from Chinese open source coding models right now. The horror house has broken game logic. Collect all three keys, complete the objective, and the front door still won't open. The stealth game shipped with lighting so dark you can't see your own character. The Remotion video is rough. GLM 5.2 is faster than 5.1 and 10x cheaper than the frontier. But Opus 4.8 and GPT 5.5 are still in a different league for serious coding work. Chinese models keep missing the attention to detail that matters. I MISS FABLE 5 ALREADY. Full review live now.

BridgeMind

24,527 views • 1 month ago

🎉 Day-0 support for in vLLM, available today in v0.23.0! Congrats to Z.ai on GLM-5.2, a flagship model built for long-horizon coding agents. ✨ 1M-token context, built to hold project-scale engineering work in a single run ✨ Tuned for long-horizon coding: large-scale implementation, automated research, and performance optimization ✨ One task can carry a full dev workflow, from requirements to a deployable product across platforms ✨ Client-side and mobile engineering, including an on-device debugging loop Try it out running it on vLLM today: 🔗

🎉 Day-0 support for in vLLM, available today in v0.23.0! Congrats to Z.ai on GLM-5.2, a flagship model built for long-horizon coding agents. ✨ 1M-token context, built to hold project-scale engineering work in a single run ✨ Tuned for long-horizon coding: large-scale implementation, automated research, and performance optimization ✨ One task can carry a full dev workflow, from requirements to a deployable product across platforms ✨ Client-side and mobile engineering, including an on-device debugging loop Try it out running it on vLLM today: 🔗

vLLM

35,925 views • 27 days ago

Excited to present the LLM-Council skill. Initial idea by Karpathy. I just packaged it as a skill. You can easily spin up a council of LLMs or agents via Fireworks AI. Watch how the new GLM-5 model "deliberates" on other LLMs' thoughts on the big question, "Can LLMs reason?" Things worth paying attention to: New open models like GLM-5 have surprisingly improved on complex reasoning and long-running agentic tasks. The AskUserQuestion tool in Claude Code came in handy to select the council and chairperson. As Andrej Karpathy puts it, it's a really interesting way to get different perspectives from LLMs, which can lead to better decision-making on whatever task you are working on. You can use it for other agentic coding use cases, like evaluation, tool building, designing, and research.

Excited to present the LLM-Council skill. Initial idea by Karpathy. I just packaged it as a skill. You can easily spin up a council of LLMs or agents via Fireworks AI. Watch how the new GLM-5 model "deliberates" on other LLMs' thoughts on the big question, "Can LLMs reason?" Things worth paying attention to: New open models like GLM-5 have surprisingly improved on complex reasoning and long-running agentic tasks. The AskUserQuestion tool in Claude Code came in handy to select the council and chairperson. As Andrej Karpathy puts it, it's a really interesting way to get different perspectives from LLMs, which can lead to better decision-making on whatever task you are working on. You can use it for other agentic coding use cases, like evaluation, tool building, designing, and research.

elvis

39,452 views • 5 months ago

Crazy... I just found a super cheap way to code with AI Tried GLM 4.5 on Claude Code and I’m really impressed! Prompt: “create an interactive python app where users can track, manage or analyze data” And, It came up with: • Dashboard - a clean interface to see all your data at a glance • Datasets - manage multiple collections of data easily • Data entries - add, edit, and organize your records without hassle • Analysis - get insights and summaries directly from your data What I observed: Everything worked smoothly, GLM 4.5 is definitely one of the best open-source models for coding, I’ll keep using it. Check it out now: And the best part: super affordable coding plans! • Lite - $3/month (promo). ~120 prompts every 5 hours. Works with Claude Code, Roo Code, Cline, Kilo Code, and more. • Pro - $15/month (promo). 5× more usage than Lite. Faster. Includes image/video tools and web search. • Max - $30/month (promo). 4× more usage than Pro. Guaranteed performance. Early access to new features. Bonus - Use my referral link to get an extra 10% off on top of these promo prices! Video demo 👇

Crazy... I just found a super cheap way to code with AI Tried GLM 4.5 on Claude Code and I’m really impressed! Prompt: “create an interactive python app where users can track, manage or analyze data” And, It came up with: • Dashboard - a clean interface to see all your data at a glance • Datasets - manage multiple collections of data easily • Data entries - add, edit, and organize your records without hassle • Analysis - get insights and summaries directly from your data What I observed: Everything worked smoothly, GLM 4.5 is definitely one of the best open-source models for coding, I’ll keep using it. Check it out now: And the best part: super affordable coding plans! • Lite - $3/month (promo). ~120 prompts every 5 hours. Works with Claude Code, Roo Code, Cline, Kilo Code, and more. • Pro - $15/month (promo). 5× more usage than Lite. Faster. Includes image/video tools and web search. • Max - $30/month (promo). 4× more usage than Pro. Guaranteed performance. Early access to new features. Bonus - Use my referral link to get an extra 10% off on top of these promo prices! Video demo 👇

Chidanand Tripathi

83,854 views • 9 months ago

People made fun of Alex Finn for buying three Mac Studios to run AI at home. Then Fable got banned for a week, GLM 5.2 dropped, and those exact Mac Studios started reselling for 4x what he paid. He showed me how he built his home AI lab from scratch. Here's the playbook: 1) The hardware. three 512GB Mac Studios, an NVIDIA DGX Spark, a custom RTX 5090 build, and a few Mac Minis. ~$30k all in. 2) The buying framework... - Mac Studio: huge memory, runs GLM 5.2 (open weights, near Opus 4.8 on benchmarks), but slow. - DGX Spark ($4,800): the sweet spot for most people. - RTX 5090: smaller models at blazing speed (Qwen's 29B now hits Sonnet 4 level). 3) Tailscale networks every machine into one private network with root access to each other. Only one machine is plugged into a monitor. 4) A Nous Research Hermes agent is his IT guy. New model drops? It SSHs into the right box, loads 5 candidates, runs evals overnight, and reports back which task belongs on which machine. Alex has literally never loaded a model himself. 5) The whole point: achieving "ambient intelligence." Always-on jobs that would bankrupt you on per-token billing. A security sweep of his API endpoints every hour. Code optimization every 20 minutes. Database anomaly & churn detection. Hourly scraping of X, Reddit & Hacker News for business opportunities. 6) Running those workloads on frontier models would cost thousands a month. His actual cost: ~$60 more in electricity. 7) Btw he's not anti-frontier. He still maxes out his Claude plan. The way he sees it: frontier is for hard thinking, local is for the foot soldiers that never sleep. 8) "We own everything except for the intelligence. Why can't we own the intelligence?" 9) He thinks frontier-level intelligence runs on consumer hardware within 6 months.

People made fun of Alex Finn for buying three Mac Studios to run AI at home. Then Fable got banned for a week, GLM 5.2 dropped, and those exact Mac Studios started reselling for 4x what he paid. He showed me how he built his home AI lab from scratch. Here's the playbook: 1) The hardware. three 512GB Mac Studios, an NVIDIA DGX Spark, a custom RTX 5090 build, and a few Mac Minis. ~$30k all in. 2) The buying framework... - Mac Studio: huge memory, runs GLM 5.2 (open weights, near Opus 4.8 on benchmarks), but slow. - DGX Spark ($4,800): the sweet spot for most people. - RTX 5090: smaller models at blazing speed (Qwen's 29B now hits Sonnet 4 level). 3) Tailscale networks every machine into one private network with root access to each other. Only one machine is plugged into a monitor. 4) A Nous Research Hermes agent is his IT guy. New model drops? It SSHs into the right box, loads 5 candidates, runs evals overnight, and reports back which task belongs on which machine. Alex has literally never loaded a model himself. 5) The whole point: achieving "ambient intelligence." Always-on jobs that would bankrupt you on per-token billing. A security sweep of his API endpoints every hour. Code optimization every 20 minutes. Database anomaly & churn detection. Hourly scraping of X, Reddit & Hacker News for business opportunities. 6) Running those workloads on frontier models would cost thousands a month. His actual cost: ~$60 more in electricity. 7) Btw he's not anti-frontier. He still maxes out his Claude plan. The way he sees it: frontier is for hard thinking, local is for the foot soldiers that never sleep. 8) "We own everything except for the intelligence. Why can't we own the intelligence?" 9) He thinks frontier-level intelligence runs on consumer hardware within 6 months.

Alex Lieberman

56,348 views • 7 days ago

Six open-source LLMs. One sliding puzzle. A brutal test of long-horizon reasoning and tool calling. Five of them broke. One didn't. I gave each model a move_tile tool and a scrambled 3×3 board, then asked it to solve the puzzle through pure turn-by-turn reasoning. The deeper the scramble, the more brutal the search. Five runs per depth, best run kept. A model fails the round if it exceeds 6x the optimal move count. > Depth 5: Everyone solves it. Yawn. > Depth 10: GLM 5.1 melts down. 43 moves. Cut. > Depth 12: Gemma4 26B loses the plot, shuffling tiles in circles. Gone. > Depth 15: The wall. DeepSeek V4 Flash, out. DeepSeek V4 Pro, out. Gemma4, out again. GLM 5.1, out. Two survivors: Qwen3.6 35B-A3B, and Kimi K2.6 with an 11-move solve that looked like cheating. > Depth 18: Same two. Everyone else hallucinating tiles that weren't there. > Depth 22: Final boss. Kimi, flawless for five rounds, finally cracks. 81 moves. Still scrambled. DeepSeek V4 Pro limps home at 90. Qwen3.6 35B-A3B solves it in 36. The smallest model in the room. ~3B active params. Fits on a single 3090. It beat everything. Kimi was elegant. Qwen3.6 was unstoppable.

Six open-source LLMs. One sliding puzzle. A brutal test of long-horizon reasoning and tool calling. Five of them broke. One didn't. I gave each model a move_tile tool and a scrambled 3×3 board, then asked it to solve the puzzle through pure turn-by-turn reasoning. The deeper the scramble, the more brutal the search. Five runs per depth, best run kept. A model fails the round if it exceeds 6x the optimal move count. > Depth 5: Everyone solves it. Yawn. > Depth 10: GLM 5.1 melts down. 43 moves. Cut. > Depth 12: Gemma4 26B loses the plot, shuffling tiles in circles. Gone. > Depth 15: The wall. DeepSeek V4 Flash, out. DeepSeek V4 Pro, out. Gemma4, out again. GLM 5.1, out. Two survivors: Qwen3.6 35B-A3B, and Kimi K2.6 with an 11-move solve that looked like cheating. > Depth 18: Same two. Everyone else hallucinating tiles that weren't there. > Depth 22: Final boss. Kimi, flawless for five rounds, finally cracks. 81 moves. Still scrambled. DeepSeek V4 Pro limps home at 90. Qwen3.6 35B-A3B solves it in 36. The smallest model in the room. ~3B active params. Fits on a single 3090. It beat everything. Kimi was elegant. Qwen3.6 was unstoppable.

stevibe

16,407 views • 2 months ago

How to setup a multi agent system? Bookmark it 📂 "The Trading Floor" Multi-Agent Market Analysis Council to analyze a stock ticker Z.ai GLM-4.7 🤝 OpenCode Agent framework: CrewAI How it works? 1. User enters a stock ticker to analyze 2. 5 AI agents wake up, each with distinct expertise: - Quant Analyst — technical indicators & price patterns - Sentiment Scout — market mood & crowd psychology - Macro Strategist — sector dynamics & economic context - Risk Manager — volatility, drawdowns & position sizing - Portfolio Chief — synthesizes all perspectives 3. Agents analyze independently using real market data 4. They debate, challenge assumptions, and identify disagreements 5. Portfolio Chief resolves conflicts and delivers a consensus recommendation 6. Final output: buy/hold/sell rating with confidence level, position size, and key risks How to built The Trading Floor? 1. Chose CrewAI as the agent framework — handles multi-agent orchestration out of the box 2. Defined 5 agents with distinct roles, goals, and backstories in Python 3. Built custom tools wrapping yfinance for real market data (prices, indicators, volatility) 4. Configured sequential workflow — specialists analyze first, Portfolio Chief synthesizes last 5. Set up FastAPI backend with SSE to stream agent thoughts in real-time 6. Built Next.js frontend to visualize the "board of directors" deliberating live 7. One environment variable (MODEL=openai/gpt-5.2) powers all agents 8. Generated unique agent icons with AI image tools Total cost: $0 for the framework, pay only for LLM API calls Tech stack: - GLM-4.7 with opencode to build the app - CrewAI (open source) for agent orchestration - GPT-5.2 powering each agent - FastAPI + SSE for real-time streaming - Next.js frontend showing live agent deliberations

How to setup a multi agent system? Bookmark it 📂 "The Trading Floor" Multi-Agent Market Analysis Council to analyze a stock ticker Z.ai GLM-4.7 🤝 OpenCode Agent framework: CrewAI How it works? 1. User enters a stock ticker to analyze 2. 5 AI agents wake up, each with distinct expertise: - Quant Analyst — technical indicators & price patterns - Sentiment Scout — market mood & crowd psychology - Macro Strategist — sector dynamics & economic context - Risk Manager — volatility, drawdowns & position sizing - Portfolio Chief — synthesizes all perspectives 3. Agents analyze independently using real market data 4. They debate, challenge assumptions, and identify disagreements 5. Portfolio Chief resolves conflicts and delivers a consensus recommendation 6. Final output: buy/hold/sell rating with confidence level, position size, and key risks How to built The Trading Floor? 1. Chose CrewAI as the agent framework — handles multi-agent orchestration out of the box 2. Defined 5 agents with distinct roles, goals, and backstories in Python 3. Built custom tools wrapping yfinance for real market data (prices, indicators, volatility) 4. Configured sequential workflow — specialists analyze first, Portfolio Chief synthesizes last 5. Set up FastAPI backend with SSE to stream agent thoughts in real-time 6. Built Next.js frontend to visualize the "board of directors" deliberating live 7. One environment variable (MODEL=openai/gpt-5.2) powers all agents 8. Generated unique agent icons with AI image tools Total cost: $0 for the framework, pay only for LLM API calls Tech stack: - GLM-4.7 with opencode to build the app - CrewAI (open source) for agent orchestration - GPT-5.2 powering each agent - FastAPI + SSE for real-time streaming - Next.js frontend showing live agent deliberations

CloudAI-X

57,703 views • 6 months ago

JARVIS-1: Open-World Multi-task Agents with Memory-Augmented Multimodal Language Models paper page: Achieving human-like planning and control with multimodal observations in an open world is a key milestone for more functional generalist agents. Existing approaches can handle certain long-horizon tasks in an open world. However, they still struggle when the number of open-world tasks could potentially be infinite and lack the capability to progressively enhance task completion as game time progresses. We introduce JARVIS-1, an open-world agent that can perceive multimodal input (visual observations and human instructions), generate sophisticated plans, and perform embodied control, all within the popular yet challenging open-world Minecraft universe. Specifically, we develop JARVIS-1 on top of pre-trained multimodal language models, which map visual observations and textual instructions to plans. The plans will be ultimately dispatched to the goal-conditioned controllers. We outfit JARVIS-1 with a multimodal memory, which facilitates planning using both pre-trained knowledge and its actual game survival experiences. In our experiments, JARVIS-1 exhibits nearly perfect performances across over 200 varying tasks from the Minecraft Universe Benchmark, ranging from entry to intermediate levels. JARVIS-1 has achieved a completion rate of 12.5% in the long-horizon diamond pickaxe task. This represents a significant increase up to 5 times compared to previous records. Furthermore, we show that JARVIS-1 is able to self-improve following a life-long learning paradigm thanks to multimodal memory, sparking a more general intelligence and improved autonomy.

JARVIS-1: Open-World Multi-task Agents with Memory-Augmented Multimodal Language Models paper page: Achieving human-like planning and control with multimodal observations in an open world is a key milestone for more functional generalist agents. Existing approaches can handle certain long-horizon tasks in an open world. However, they still struggle when the number of open-world tasks could potentially be infinite and lack the capability to progressively enhance task completion as game time progresses. We introduce JARVIS-1, an open-world agent that can perceive multimodal input (visual observations and human instructions), generate sophisticated plans, and perform embodied control, all within the popular yet challenging open-world Minecraft universe. Specifically, we develop JARVIS-1 on top of pre-trained multimodal language models, which map visual observations and textual instructions to plans. The plans will be ultimately dispatched to the goal-conditioned controllers. We outfit JARVIS-1 with a multimodal memory, which facilitates planning using both pre-trained knowledge and its actual game survival experiences. In our experiments, JARVIS-1 exhibits nearly perfect performances across over 200 varying tasks from the Minecraft Universe Benchmark, ranging from entry to intermediate levels. JARVIS-1 has achieved a completion rate of 12.5% in the long-horizon diamond pickaxe task. This represents a significant increase up to 5 times compared to previous records. Furthermore, we show that JARVIS-1 is able to self-improve following a life-long learning paradigm thanks to multimodal memory, sparking a more general intelligence and improved autonomy.

AK

141,425 views • 2 years ago

Which LLM reasons best when it doesn't have all the information? Enter LLM Poker Arena to find out. It's a Poker Playing benchmark where top reasoning models play Texas Hold'em poker against each other. Claude Opus 4.5, GPT-5.2, Gemini 2.5 Pro, and Grok 4 all sit at the same table and play full tournaments to see who finishes with the chips. Poker is very different when it comes to reasoning. It has to balance probabilistic reasoning, opponent modeling and make decisions under uncertainty. Poker is an interesting evaluation because it tests reasoning under incomplete information, something most coding benchmarks do not capture. In this tournaments the rules are: - Each LLM starts with $1,000 chips - Small and big blinds start at $25 / $50 - Blinds double every 3 minutes - All models run in their reasoning or thinking modes After the first 5 tournaments: - Claude Opus 4.5 with Thinking has 3 wins - GPT-5.2 has 2 wins - Grok 4 and Gemini 2.5 Pro have 0 wins Early results suggest Claude performs quite well at poker as well. Also five is a very small sample size. Planning to run many more tournaments, publish the full benchmark data and add a prediction market on top of it. Thanks for the suggestion clipz. Much more coming as part of Poker Cities !! This was built on Replit ⠕ using their AI integrations, which made it straightforward to connect Claude, GPT, and Gemini. What model do you think wins after 100 tournaments?

Which LLM reasons best when it doesn't have all the information? Enter LLM Poker Arena to find out. It's a Poker Playing benchmark where top reasoning models play Texas Hold'em poker against each other. Claude Opus 4.5, GPT-5.2, Gemini 2.5 Pro, and Grok 4 all sit at the same table and play full tournaments to see who finishes with the chips. Poker is very different when it comes to reasoning. It has to balance probabilistic reasoning, opponent modeling and make decisions under uncertainty. Poker is an interesting evaluation because it tests reasoning under incomplete information, something most coding benchmarks do not capture. In this tournaments the rules are: - Each LLM starts with $1,000 chips - Small and big blinds start at $25 / $50 - Blinds double every 3 minutes - All models run in their reasoning or thinking modes After the first 5 tournaments: - Claude Opus 4.5 with Thinking has 3 wins - GPT-5.2 has 2 wins - Grok 4 and Gemini 2.5 Pro have 0 wins Early results suggest Claude performs quite well at poker as well. Also five is a very small sample size. Planning to run many more tournaments, publish the full benchmark data and add a prediction market on top of it. Thanks for the suggestion clipz. Much more coming as part of Poker Cities !! This was built on Replit ⠕ using their AI integrations, which made it straightforward to connect Claude, GPT, and Gemini. What model do you think wins after 100 tournaments?

Anshul Dhawan

32,192 views • 5 months ago

This is the moment Chinese AI beat American AI. One of the largest public crypto companies in the world just DUMPED OpenAI and Anthropic. Coinbase switched to open-weight Chinese models from Zhipu and DeepSeek, and shaved nearly 50% off the company's internal AI spending. The numbers are absolutely ridiculous: Running the same enterprise workload through Anthropic's Claude costs $4,811. Running it through Zhipu's GLM 5.2 costs $544. That's a 9x price difference for equivalent output. OpenAI's GPT-5.5 sits in the middle at $3,357. DeepSeek's V4 lands at $1,071. Moonshot's Kimi at $948. On the actual benchmarks: Zhipu's GLM 5.2 scored 62.1 on SWE-bench Pro, the gold standard for coding. OpenAI's GPT-5.5 scored 58.6. One AI researcher called GLM 5.2 "at least as good as Opus 4.8 and GPT 5.5." Another called it "the first open model that can really compete with closed-source systems." The Chinese models are not just cheaper but they are now also beating American models on the benchmarks American companies pay $4,811 per workload for. Coinbase did the math first and reacted - more companies will certainly follow. Now watch what happens to the IPO timeline: Anthropic confidentially filed for an IPO targeting October at a $965 billion valuation. OpenAI followed days later with its own confidential filing. Both companies built their financial models on the assumption that they could keep charging enterprise prices that are 9 to 33x what Chinese competitors charge for the same task. Brian Armstrong publicly proved customers WILL leave. 45% of companies are now spending over $100,000 per month on AI, up from 20% last year. Every one of those customers is one quarterly budget review away from dumping American AI. OpenAI has reportedly already started preparing major token price cuts. Anthropic is expected to follow. And here's the thing... The export controls were supposed to CRUSH Chinese AI. The US government banned American AI chips, restricted model weights, blacklisted Alibaba and Baidu as Chinese military companies, and just banned Anthropic's flagship model from every foreign national on the planet. The entire premise of the American AI valuation bubble is that Washington can keep China two generations behind. But Chinese labs responded by building cheaper, more efficient models on inferior hardware and pricing them at one ninth the cost of the American alternative. And now American companies are voting with their checkbooks. The dominant American labs are valued at nearly $2 trillion combined on the assumption that their pricing power is durable. Coinbase proved it is not, and every customer doing a year-end budget review will be looking at the same math. For investors, the question here is what happens to the Anthropic IPO at $965 billion when the company is being forced to cut prices to defend share against open-weight Chinese models that score higher on the benchmarks. For everyone else, the bigger question is what happens when Washington spent four years and billions of dollars trying to contain Chinese AI, and the only thing that actually shifted in the end was American customers.

This is the moment Chinese AI beat American AI. One of the largest public crypto companies in the world just DUMPED OpenAI and Anthropic. Coinbase switched to open-weight Chinese models from Zhipu and DeepSeek, and shaved nearly 50% off the company's internal AI spending. The numbers are absolutely ridiculous: Running the same enterprise workload through Anthropic's Claude costs $4,811. Running it through Zhipu's GLM 5.2 costs $544. That's a 9x price difference for equivalent output. OpenAI's GPT-5.5 sits in the middle at $3,357. DeepSeek's V4 lands at $1,071. Moonshot's Kimi at $948. On the actual benchmarks: Zhipu's GLM 5.2 scored 62.1 on SWE-bench Pro, the gold standard for coding. OpenAI's GPT-5.5 scored 58.6. One AI researcher called GLM 5.2 "at least as good as Opus 4.8 and GPT 5.5." Another called it "the first open model that can really compete with closed-source systems." The Chinese models are not just cheaper but they are now also beating American models on the benchmarks American companies pay $4,811 per workload for. Coinbase did the math first and reacted - more companies will certainly follow. Now watch what happens to the IPO timeline: Anthropic confidentially filed for an IPO targeting October at a $965 billion valuation. OpenAI followed days later with its own confidential filing. Both companies built their financial models on the assumption that they could keep charging enterprise prices that are 9 to 33x what Chinese competitors charge for the same task. Brian Armstrong publicly proved customers WILL leave. 45% of companies are now spending over $100,000 per month on AI, up from 20% last year. Every one of those customers is one quarterly budget review away from dumping American AI. OpenAI has reportedly already started preparing major token price cuts. Anthropic is expected to follow. And here's the thing... The export controls were supposed to CRUSH Chinese AI. The US government banned American AI chips, restricted model weights, blacklisted Alibaba and Baidu as Chinese military companies, and just banned Anthropic's flagship model from every foreign national on the planet. The entire premise of the American AI valuation bubble is that Washington can keep China two generations behind. But Chinese labs responded by building cheaper, more efficient models on inferior hardware and pricing them at one ninth the cost of the American alternative. And now American companies are voting with their checkbooks. The dominant American labs are valued at nearly $2 trillion combined on the assumption that their pricing power is durable. Coinbase proved it is not, and every customer doing a year-end budget review will be looking at the same math. For investors, the question here is what happens to the Anthropic IPO at $965 billion when the company is being forced to cut prices to defend share against open-weight Chinese models that score higher on the benchmarks. For everyone else, the bigger question is what happens when Washington spent four years and billions of dollars trying to contain Chinese AI, and the only thing that actually shifted in the end was American customers.

Ricardo

249,975 views • 16 days ago

One of the things I’m most excited about this year is building agents that can work productively for hours, days, or weeks. Coding agents are starting to become very competent at this, but what about computer use agents? Our new benchmark, Odysseys (co-led with Lawrence Jang) is a set of 200 new tasks derived from real world browsing behavior that measure long horizon web navigation capabilities (potentially up to hours of web browsing work). Interestingly, we find that frontier CUAs are already surprisingly good at working productively for up to an hour on these tasks, but there’s a lot of work to be done in making them even more efficient. Like every other AI researcher, my real dream is to open a cafe once we solve ASI. So, here’s Opus 4.6 doing some market research for me ("I want to do market research on the most popular cafes in Singapore. Analyse the menus of the top 10 cafes in Singapore (by Google reviews/ratings), and make sure we include at least 1 from the North/South/East/West/Central regions of Singapore. Keep the relevant pages of each cafe open, and summarise their pricing, menu offerings, unique selling points, making sure to reference which tab is opened for each cafe. For each cafe, also help me figure out how long it would take to get to it from Tampines MRT, and include this in your final summary."). I was very impressed to see Opus 4.6 complete this task after working for 52 mins, satisfying all 7 rubrics that corresponded to this task. It provided a very nice markdown summary at the end that gave me all the information I asked for!

One of the things I’m most excited about this year is building agents that can work productively for hours, days, or weeks. Coding agents are starting to become very competent at this, but what about computer use agents? Our new benchmark, Odysseys (co-led with Lawrence Jang) is a set of 200 new tasks derived from real world browsing behavior that measure long horizon web navigation capabilities (potentially up to hours of web browsing work). Interestingly, we find that frontier CUAs are already surprisingly good at working productively for up to an hour on these tasks, but there’s a lot of work to be done in making them even more efficient. Like every other AI researcher, my real dream is to open a cafe once we solve ASI. So, here’s Opus 4.6 doing some market research for me ("I want to do market research on the most popular cafes in Singapore. Analyse the menus of the top 10 cafes in Singapore (by Google reviews/ratings), and make sure we include at least 1 from the North/South/East/West/Central regions of Singapore. Keep the relevant pages of each cafe open, and summarise their pricing, menu offerings, unique selling points, making sure to reference which tab is opened for each cafe. For each cafe, also help me figure out how long it would take to get to it from Tampines MRT, and include this in your final summary."). I was very impressed to see Opus 4.6 complete this task after working for 52 mins, satisfying all 7 rubrics that corresponded to this task. It provided a very nice markdown summary at the end that gave me all the information I asked for!

Jing Yu Koh

49,418 views • 2 months ago

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.

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 views • 1 year ago

CHINA JUST DROPPED AN AI CODING MODEL WITH A 1M CONTEXT WINDOW. And I connected it to Claude Code to see what it could actually do. Meet GLM-X Preview On paper, a few things immediately stood out: → 1M context window → Agentic coding capabilities → Works inside Claude Code → Designed for large-scale coding and reasoning workflows But specs don't matter much if the model can't deliver in practice. So I gave it a real-world task. THE TEST One prompt: > Build a modern AI lead generation dashboard using React and Tailwind CSS. Requirements: → Dark mode → Analytics dashboard → Lead table → Email outreach section → Responsive design → Production-ready component structure Instead of generating a few snippets, it planned the architecture, generated the dashboard components, created the Tailwind configuration, and walked through the implementation requirements. What impressed me most wasn't the code itself. It was how well it maintained context throughout the workflow. That's where a 1M context window starts becoming useful. Less time re-explaining requirements. Less context loss. More room for complex projects. The AI coding race is getting very interesting. And it's no longer just GPT, Claude, and Gemini competing for attention. Results from my test below 👇

CHINA JUST DROPPED AN AI CODING MODEL WITH A 1M CONTEXT WINDOW. And I connected it to Claude Code to see what it could actually do. Meet GLM-X Preview On paper, a few things immediately stood out: → 1M context window → Agentic coding capabilities → Works inside Claude Code → Designed for large-scale coding and reasoning workflows But specs don't matter much if the model can't deliver in practice. So I gave it a real-world task. THE TEST One prompt: > Build a modern AI lead generation dashboard using React and Tailwind CSS. Requirements: → Dark mode → Analytics dashboard → Lead table → Email outreach section → Responsive design → Production-ready component structure Instead of generating a few snippets, it planned the architecture, generated the dashboard components, created the Tailwind configuration, and walked through the implementation requirements. What impressed me most wasn't the code itself. It was how well it maintained context throughout the workflow. That's where a 1M context window starts becoming useful. Less time re-explaining requirements. Less context loss. More room for complex projects. The AI coding race is getting very interesting. And it's no longer just GPT, Claude, and Gemini competing for attention. Results from my test below 👇

Md Riyazuddin

31,199 views • 27 days ago

For developers, Gemini 3 Pro offers unparalleled results across every major AI benchmark compared to Gemini 2.5 Pro. Here three ways Gemini 3 Pro can improve your workflow today: 1. Gemini 3 Pro fits right into existing production agent and coding workflows. It is also integrated into development environment experiences across the ecosystem of developer products like Google AI Studio, Gemini CLI, and Android Studio. 2. We introduced Google Google Antigravity, a new agentic development platform that reimagines how the model and IDE work together. It enables developers to manage agents to work at a higher, task-oriented level collaborating with agents while they build. 3. We’ve built Google AI Studio to be your fastest path from a prompt to an AI-native app. Build mode lets you add AI capabilities faster than ever, automatically wiring up the right models and APIs, while features like annotations enable fast and intuitive iteration. Whether you’re building a one-shot prompt game or an interactive landing page from unstructured voice notes, you can bring any idea to life with Gemini 3.

For developers, Gemini 3 Pro offers unparalleled results across every major AI benchmark compared to Gemini 2.5 Pro. Here three ways Gemini 3 Pro can improve your workflow today: 1. Gemini 3 Pro fits right into existing production agent and coding workflows. It is also integrated into development environment experiences across the ecosystem of developer products like Google AI Studio, Gemini CLI, and Android Studio. 2. We introduced Google Google Antigravity, a new agentic development platform that reimagines how the model and IDE work together. It enables developers to manage agents to work at a higher, task-oriented level collaborating with agents while they build. 3. We’ve built Google AI Studio to be your fastest path from a prompt to an AI-native app. Build mode lets you add AI capabilities faster than ever, automatically wiring up the right models and APIs, while features like annotations enable fast and intuitive iteration. Whether you’re building a one-shot prompt game or an interactive landing page from unstructured voice notes, you can bring any idea to life with Gemini 3.

Google AI

136,589 views • 7 months ago

ChatGPT: Opens a new tab, you copy-paste. Siri: Transcribes text, you still do the work. Lemon: Completes the entire task while you keep working. Traditional voice assistants are convenient to use. But they cannot complete tasks for you. That's why Lemon stands out. Lemon is an AI agent that completes tasks from voice commands, without opening new tabs or switching apps. Press the fn key. Say what you need done. Watch it happen: ➟ "Reply to Sarah's email and tell her I'll have the report by Friday" — sent ➟ "Create a project brief for the new client" — done ➟ "Search for the latest stats on AI adoption" — on your screen ➟ "Make this email sound more professional" — edited ➟ "Give me three angles for this campaign" — ideas generated ➟ "Write this paragraph in a more casual tone" — rewritten No new tabs. No manual steps. No switching back and forth. Lemon works across every app on your Mac. The results: ✓ Users save 2+ hours daily ✓ Reply to emails 12x faster ✓ Open 70% fewer tabs ✓ Type 5x faster than traditional workflows Your voice isn't just for dictation anymore. It's a faster way to complete your work tasks. 📌 Try Lemon (it's free for 30 days): 🔄 Repost to help others work faster — #AI #Agent #LemonPartnership

ChatGPT: Opens a new tab, you copy-paste. Siri: Transcribes text, you still do the work. Lemon: Completes the entire task while you keep working. Traditional voice assistants are convenient to use. But they cannot complete tasks for you. That's why Lemon stands out. Lemon is an AI agent that completes tasks from voice commands, without opening new tabs or switching apps. Press the fn key. Say what you need done. Watch it happen: ➟ "Reply to Sarah's email and tell her I'll have the report by Friday" — sent ➟ "Create a project brief for the new client" — done ➟ "Search for the latest stats on AI adoption" — on your screen ➟ "Make this email sound more professional" — edited ➟ "Give me three angles for this campaign" — ideas generated ➟ "Write this paragraph in a more casual tone" — rewritten No new tabs. No manual steps. No switching back and forth. Lemon works across every app on your Mac. The results: ✓ Users save 2+ hours daily ✓ Reply to emails 12x faster ✓ Open 70% fewer tabs ✓ Type 5x faster than traditional workflows Your voice isn't just for dictation anymore. It's a faster way to complete your work tasks. 📌 Try Lemon (it's free for 30 days): 🔄 Repost to help others work faster — #AI #Agent #LemonPartnership

Andrew Bolis

91,782 views • 4 months ago

China just released an open source AI model that matches the best closed models from OpenAI and Anthropic. Gavin Baker explained exactly how they did it and the answer should concern every American AI lab. The model is called GLM 5.2. It was built by Z. AI. You get 744 billion parameters, 1 million token context window and its MIT license, meaning anyone can download it, fork it, build a company on it, with no restrictions and no Dario. It scored 51 points on the artificial analysis intelligence index. The highest score any open weight model has ever achieved. It beat GPT 5.5 on the frontier software engineering benchmark. It trails Claude Opus 4.8 by less than one percentage point. And it costs 85% less to run than GPT 5.5 for comparable performance. Gavin Baker said on the All-In podcast that this model has challenged some of his beliefs. Then he explained how China built it. The method is called distillation. Just think of tens of thousands of phones and computers running simultaneously, all hitting the frontier model APIs through masked accounts, asking specific questions, and harvesting what happens inside the model when it answers. Every reasoning step, every token. The entire thinking process gets recorded and fed back into the Chinese model during training. It is a cheat sheet. It is the answer key to the exam. And here is the part that should worry everyone. Sacks said it plainly. China was already nine months behind American models. But now that GLM 5.2 is good enough to run its own reinforcement learning, it can improve itself without needing to distill from American models anymore. The cheat sheet let them get close enough to start writing their own answers. Sacks said we are six months behind on the model and 24 months behind on silicon and they are only a few months behind in total. The Z. AI founder told Elon Musk directly that open weight fable-level capability will be here before Q1 2027. Every restriction Anthropic lobbied for, every self-imposed safety guardrail, every month of delay in releasing American frontier models accelerated this. The Chinese labs were not under those restrictions. They were not going to wait. The composable model future Gavin described, where every enterprise runs a frontier model alongside their own fine-tuned open weight model, is coming regardless of what American labs do next. The question is just whether the open weight half of that stack is American or Chinese. Right now it is Chinese. WATCH THE FULL PODCAST ON The All-In Podcast

China just released an open source AI model that matches the best closed models from OpenAI and Anthropic. Gavin Baker explained exactly how they did it and the answer should concern every American AI lab. The model is called GLM 5.2. It was built by Z. AI. You get 744 billion parameters, 1 million token context window and its MIT license, meaning anyone can download it, fork it, build a company on it, with no restrictions and no Dario. It scored 51 points on the artificial analysis intelligence index. The highest score any open weight model has ever achieved. It beat GPT 5.5 on the frontier software engineering benchmark. It trails Claude Opus 4.8 by less than one percentage point. And it costs 85% less to run than GPT 5.5 for comparable performance. Gavin Baker said on the All-In podcast that this model has challenged some of his beliefs. Then he explained how China built it. The method is called distillation. Just think of tens of thousands of phones and computers running simultaneously, all hitting the frontier model APIs through masked accounts, asking specific questions, and harvesting what happens inside the model when it answers. Every reasoning step, every token. The entire thinking process gets recorded and fed back into the Chinese model during training. It is a cheat sheet. It is the answer key to the exam. And here is the part that should worry everyone. Sacks said it plainly. China was already nine months behind American models. But now that GLM 5.2 is good enough to run its own reinforcement learning, it can improve itself without needing to distill from American models anymore. The cheat sheet let them get close enough to start writing their own answers. Sacks said we are six months behind on the model and 24 months behind on silicon and they are only a few months behind in total. The Z. AI founder told Elon Musk directly that open weight fable-level capability will be here before Q1 2027. Every restriction Anthropic lobbied for, every self-imposed safety guardrail, every month of delay in releasing American frontier models accelerated this. The Chinese labs were not under those restrictions. They were not going to wait. The composable model future Gavin described, where every enterprise runs a frontier model alongside their own fine-tuned open weight model, is coming regardless of what American labs do next. The question is just whether the open weight half of that stack is American or Chinese. Right now it is Chinese. WATCH THE FULL PODCAST ON The All-In Podcast

Ihtesham Ali

85,915 views • 17 days ago

Cerebras inference is very fast. So fast that it changes how we think about configuring our LLMs for voice agent use cases. Kimi K2.6 is a 1T parameter reasoning model that Cerebras serves at 650 - 1,000 tokens per second (end-to-end throughput), with time to first token metrics as low as 150ms (latency). These numbers are two to three times faster than other similarly capable models. The biggest lever we get from this kind of speed is that we can use the model in reasoning mode, and still have excellent "time to first non-thinking token." This solves a big pain point we have in 2026 for voice agent use cases. Almost all recent innovation in post-training has focused on making models good at reasoning ("test time compute"). This is great, but it makes the user-facing model latency much, much slower. Which is a problem for conversational voice agents. We can run Kimi K2.6 with reasoning turned on, and get responses faster than other models produce with reasoning disabled. On my 30-turn voice agent benchmark, Kimi K2.6 with reasoning enabled ties GPT 5.1 and Haiku 4.5 with reasoning disabled, and is still about 200ms seconds faster! On my primary task agent benchmark, Kimi K2.6 is now the #2 model. It ranks just behind Gemini 3.5 Flash in "high" reasoning mode, and tied with GLM 5, Sonnet 4.6, and GPT 5.4 with reasoning set to "low." But Kimi K2.6 completes each turn in the agent loop in under 500ms. The other four models are all at least 3x slower. (Models only qualify for this benchmark if they can complete task turns at a P50 <4s.) A couple of other things that this speed buys us, for production voice agents: - Tool calls happen fast enough that we don't have to work around tool call latency in our pipeline design. - We can prompt the model to output structured data at the beginning of a response, followed by plain text for voice generation. This opens up possibilities like asking the model to do complex classification/generation tasks that influence the rest of the pipeline. For example, the model could create a detailed style prompt for a steerable TTS model, for each individual conversation turn. And, of course, you can use Kimi K2.6 with reasoning turned off. Cerebras calls this "instant" mode. Here's a video of a Cerebras Kimi K2.6 voice agent with voice-to-voice response time, measured at the client, under 500ms. This is the true response latency as perceived by the user, including all network and audio codec overhead, transcription and turn detection, Kimi K2.6 token generation, and voice generation. 500ms is, effectively, instant. So the Cerebras naming for this mode is a propos. :-)

Cerebras inference is very fast. So fast that it changes how we think about configuring our LLMs for voice agent use cases. Kimi K2.6 is a 1T parameter reasoning model that Cerebras serves at 650 - 1,000 tokens per second (end-to-end throughput), with time to first token metrics as low as 150ms (latency). These numbers are two to three times faster than other similarly capable models. The biggest lever we get from this kind of speed is that we can use the model in reasoning mode, and still have excellent "time to first non-thinking token." This solves a big pain point we have in 2026 for voice agent use cases. Almost all recent innovation in post-training has focused on making models good at reasoning ("test time compute"). This is great, but it makes the user-facing model latency much, much slower. Which is a problem for conversational voice agents. We can run Kimi K2.6 with reasoning turned on, and get responses faster than other models produce with reasoning disabled. On my 30-turn voice agent benchmark, Kimi K2.6 with reasoning enabled ties GPT 5.1 and Haiku 4.5 with reasoning disabled, and is still about 200ms seconds faster! On my primary task agent benchmark, Kimi K2.6 is now the #2 model. It ranks just behind Gemini 3.5 Flash in "high" reasoning mode, and tied with GLM 5, Sonnet 4.6, and GPT 5.4 with reasoning set to "low." But Kimi K2.6 completes each turn in the agent loop in under 500ms. The other four models are all at least 3x slower. (Models only qualify for this benchmark if they can complete task turns at a P50 <4s.) A couple of other things that this speed buys us, for production voice agents: - Tool calls happen fast enough that we don't have to work around tool call latency in our pipeline design. - We can prompt the model to output structured data at the beginning of a response, followed by plain text for voice generation. This opens up possibilities like asking the model to do complex classification/generation tasks that influence the rest of the pipeline. For example, the model could create a detailed style prompt for a steerable TTS model, for each individual conversation turn. And, of course, you can use Kimi K2.6 with reasoning turned off. Cerebras calls this "instant" mode. Here's a video of a Cerebras Kimi K2.6 voice agent with voice-to-voice response time, measured at the client, under 500ms. This is the true response latency as perceived by the user, including all network and audio codec overhead, transcription and turn detection, Kimi K2.6 token generation, and voice generation. 500ms is, effectively, instant. So the Cerebras naming for this mode is a propos. :-)

kwindla

40,319 views • 1 month ago

Chamath is making one of the most important business arguments of 2026. Half of large US companies right now cannot generate returns that exceed their cost of capital, which has normalized back to its long run average of 8 to 11%. Another one in seven companies globally is stuck generating persistent returns between 1 and 5% and most businesses don't have room for error and in this environment walks every frontier AI lab saying the same thing, give us your data, your workflows, your processes and our model will make everything better. And companies by the millions said yes. What they didn't fully account for is what happens on the other side of that door. Every time an employee runs a query through a frontier model API, the prompt goes through external servers, workflows, customer data, pricing logic, internal processes, all of it transmitted through a third party. As Alex Karp said companies are spending on tokens while handing over the exact proprietary advantages that make their business worth owning. Microsoft blocked internal use of Anthropic's Claude Fable 5 but over its 30-day data retention policy and the largest software company in the world decided a frontier model's data handling was too risky for its own employees. A US government action revoked access to another frontier model for foreign nationals overnight. Now here's where the cost math becomes impossible to ignore. Deutsche Bank calculated a roughly 65x cost gap between frontier models like Claude Fable 5 at ~$3.25 per task and open-source alternatives at ~$0.05. For 90% of everyday enterprise tasks, performance is comparable. Open-weight models now match closed frontier systems on core agent tasks at roughly one-tenth the cost, a high-volume deployment that costs $250/day on Claude runs at $12/day on an open-source equivalent. Chamath Palihapitiya tested this directly by running a standard enterprise code migration task through an orchestration layer wrapping an open-source model came in 16.4x cheaper than using a frontier model directly.

Chamath is making one of the most important business arguments of 2026. Half of large US companies right now cannot generate returns that exceed their cost of capital, which has normalized back to its long run average of 8 to 11%. Another one in seven companies globally is stuck generating persistent returns between 1 and 5% and most businesses don't have room for error and in this environment walks every frontier AI lab saying the same thing, give us your data, your workflows, your processes and our model will make everything better. And companies by the millions said yes. What they didn't fully account for is what happens on the other side of that door. Every time an employee runs a query through a frontier model API, the prompt goes through external servers, workflows, customer data, pricing logic, internal processes, all of it transmitted through a third party. As Alex Karp said companies are spending on tokens while handing over the exact proprietary advantages that make their business worth owning. Microsoft blocked internal use of Anthropic's Claude Fable 5 but over its 30-day data retention policy and the largest software company in the world decided a frontier model's data handling was too risky for its own employees. A US government action revoked access to another frontier model for foreign nationals overnight. Now here's where the cost math becomes impossible to ignore. Deutsche Bank calculated a roughly 65x cost gap between frontier models like Claude Fable 5 at ~$3.25 per task and open-source alternatives at ~$0.05. For 90% of everyday enterprise tasks, performance is comparable. Open-weight models now match closed frontier systems on core agent tasks at roughly one-tenth the cost, a high-volume deployment that costs $250/day on Claude runs at $12/day on an open-source equivalent. Chamath Palihapitiya tested this directly by running a standard enterprise code migration task through an orchestration layer wrapping an open-source model came in 16.4x cheaper than using a frontier model directly.

Milk Road AI

279,448 views • 10 days ago

here's what i vibecoded today: punchingface 🥊 an app to make Hugging Face models fight each other on coding and canvas challenges, built with qwen3.6 35b a3b in 24 hours! benchmarks numbers don't mean anything anymore, we need a way to visualize what the models are actually capable of, and canvas are one the best way to showcase it imo. why? because one single error in the code and everything breaks it shows the differences in a matter of seconds, way easier than manually reviewing the quality of the code produced for a complex project; much needed in the space with all the new finetunes dropping everyday! i did a little demo here with qwen3.6 vs qwopus glm 18b merged, the frankenstein model from Kyle Hessling the winner is clear here, qwen is crazy good and has nothing to prove. that said, qwopus 18b isn't terrible at all; the result isn’t the prettiest to the eye, but hey… it works! i've seen so many models just output a blank page (completely non-working code) so this is already a win frankenstein talks and thinks but he needs some extra brain surgery 🧠 results were expected (it's a very experimental model) but love the effort in the 18b direction from jackrong and kyle! the app was entirely vibecoded with qwen3.6, i didn't edit a single file manually. i can say with confidence that it really has the intelligence of claude sonnet 4.5 at a speed of 125tok/s on an rtx 5080 which models should i make fight next?

here's what i vibecoded today: punchingface 🥊 an app to make Hugging Face models fight each other on coding and canvas challenges, built with qwen3.6 35b a3b in 24 hours! benchmarks numbers don't mean anything anymore, we need a way to visualize what the models are actually capable of, and canvas are one the best way to showcase it imo. why? because one single error in the code and everything breaks it shows the differences in a matter of seconds, way easier than manually reviewing the quality of the code produced for a complex project; much needed in the space with all the new finetunes dropping everyday! i did a little demo here with qwen3.6 vs qwopus glm 18b merged, the frankenstein model from Kyle Hessling the winner is clear here, qwen is crazy good and has nothing to prove. that said, qwopus 18b isn't terrible at all; the result isn’t the prettiest to the eye, but hey… it works! i've seen so many models just output a blank page (completely non-working code) so this is already a win frankenstein talks and thinks but he needs some extra brain surgery 🧠 results were expected (it's a very experimental model) but love the effort in the 18b direction from jackrong and kyle! the app was entirely vibecoded with qwen3.6, i didn't edit a single file manually. i can say with confidence that it really has the intelligence of claude sonnet 4.5 at a speed of 125tok/s on an rtx 5080 which models should i make fight next?

left curve dev

19,111 views • 2 months ago