Microsoft at Ignite: you don't need the fanciest model for most jobs.
Their new coding model scores 98% of GPT-6.1 Sol's at about 40% of the price. It's already in Copilot.
Everyone else sells the genius. Microsoft sells the cheap, reliable mate.
microsoft.ai/news/microsoft-…
26
Meetings are becoming the build step.
Claude now joins your Google Meet and drafts the doc, whiteboard and plan while you're still talking. Beta on Team and Enterprise.
AI notes were never the point. Skipping the work after the call is.
claude.com/resources/article…
63
The number I care about on Anthropic's IPO page isn't the $17B quarter. It's "over 4x compute efficiency gain since Q1'25".
Revenue per dollar of compute is the real moat in this race. Pricing Nov 18, trading as ANT on Nov 19.
ipo.anthropic.com/us/
9
A US lab put a 65B dense model on Hugging Face that beats half-trillion open weights on agentic coding.
Ripple-65B Preview. $32M full R&D, not just the final run. Apache 2.0. Self-reported 80.5 on Terminal-Bench 2.1.
Scale is getting less interesting than taste.
ricursive.com/ripple
22
Robot AI has a demo problem. Everyone posts the one clip that worked.
RoboColosseum puts 100+ real robots in 140+ real scenes and publishes every rollout, failures included. 10,000 free eval hours too.
Publishing the failures is the whole point. That's how you find out what actually works.
teapawt.org/RoboColosseum/status/2…
Introducing RoboColosseum
An open robot eval initiative by @huggingface @BitRobotNetwork @eaic_sg @DJiafei
Access:
- 100+ robots (@UnitreeRobotics G1s, @LeRobotHF SO-101s, YAMs, Franka and more)
- 140+ in-the-wild scenes
- Open rollout data
- 10,000 FREE eval hrs
Thread ↓
1
15
Enterprise AI spend went from $33B to $177B in one year (Menlo Ventures). Agents now burn ~80% of tokens on OpenRouter.
Nobody's piloting anymore. If your product can't let an agent finish the job end to end, you're the budget line that gets cut next.
teapawt.org/MenloVentures/status/2…
This just in: Menlo Ventures’ 2026 The State of Enterprise AI report is live.
In 2026, enterprise AI spend skyrocketed 5.4x from $33B to $177B in a single year.
The full report covers how AI became an economy-scale force this year, and shows where the dollars are going, which markets are working, and which models buyers choose.
More from @timt, @mejoff, @deedydas, and @derekgxiao: mnlo.vc/enterprise-ai-26
23
Revolut moved ~2,000 engineers to self-hosted GLM-5.3 on 56 B200s. $0.30/M tokens, about a tenth of their old Anthropic/OpenAI bill.
Best bit: lead time fell more than coding time. Stop rationing tokens and people hand AI the Jira and inbox work too.
youtube.com/watch?v=Ok6QdZHC…
1
39
A food delivery company just shipped a 1.6T-parameter multimodal agent model. Meituan's LongCat-2.5 (~48B active) is built for long-horizon work: whole codebases, screenshot-to-code, multi-step tasks.
Frontier-scale agents aren't a four-lab club anymore. Anyone with real workflows to train on is in the game.
teapawt.org/Meituan_LongCat/status…
Introducing LongCat-2.5, our latest natively multimodal agentic model with 1.6T total parameters and ~48B active.
Built for long-horizon agentic tasks, it brings stronger visual understanding and agentic coding to work ranging from complex coding projects to visual coding such as creative design and 3D tasks.
Learn more about LongCat-2.5: longcat.ai/blog/longcat-2.5/
36
Glif just launched an "AI modeling agency": pick an AI model, drop in your product, get the whole ad back.
Cool demo, but the moat isn't the faces. It's keeping one character identical across 50 shots and 6 months of campaigns. That's the hard part nobody shows.
teapawt.org/heyglif/status/2116198…
Today we’re launching the first AI modeling agency 💅 right inside Glif.
Cast hundreds of real AI models, drop your product, idea or brief and let Glif create super compelling content for you end to end.
If you still think AI models aren’t ready for real work, watch this 👇
24
The largest proof in maths history, ~15,000 pages from 100+ mathematicians over decades, is now machine-checked in Lean.
Axiom's agent swarm did it in about 24 days.
A year ago agents were writing 100-line proofs. Lesson for builders: agents that keep going until the job is done beat clever one-shots.
teapawt.org/axiommathai/status/211…
The largest math proof in human history is now computer-verified.
Today, Axiom announces a complete Lean auto-formalization of the Classification of Finite Simple Groups, the "periodic table of symmetry."
About 15,000 pages, built by over 100 mathematicians across generations.
1
27
GPT-6.1 Terra is out: computer use at $1/$5 per million tokens, half or less of what GPT-5.6 Terra cost.
For anyone running agents in loops, that's the number that matters. My agents rarely fail because the model is dumb. They fail because the run gets too expensive to let it finish.
Cheap and good enough at screen work beats smart and pricey for most jobs.
teapawt.org/OpenAI/status/21147807…
GPT-6.1 Terra: delivering fast and capable computer use at a lower price.
GPT-6.1 Terra understands what’s on screen, takes action across apps, writes naturally, and is a more capable model for your whole workday with the efficiency to take on more of it.
2
35
Jensen at GTC Berlin: Nvidia's agent software should cost about $2,000 per GPU per year. No product or date yet.
Hardware margin was the moat. Now they want rent on the software stack too. Problem: vLLM and SGLang already do the inference layer for free. They'll need to sell something open source can't.
techspot.com/news/114333-nvi…
1
23
Thariq is leaving Anthropic after a year+ on Claude Code.
That tool went from curiosity to something I open every single day. A lot of why: the team shipped fast and actually listened to people on here.
Whoever picks it up next has a high bar. Thanks for the work.
teapawt.org/trq212/status/21138440…
I’ve left Anthropic after more than a year of working on Claude Code and seeing it scale larger than I ever imagined.
I have really loved working at Ant and believe deeply in the mission, and most of all, the idea of a race to the top.
The idea that if we work hard and responsibly, we can raise the bar and inspire others to great outcomes. And if we all push each other to be better, things can be radically good. I will carry the spirit of this with me forever.
My colleagues at Anthropic are the most talented, thoughtful, competent, and earnest people I have ever met. I am glad they are at the wheel and trust them deeply.
To the community on X, thank you for everything. You’ve made this the most fun (and sometimes stressful) job I’ve ever had.
For me, it is time for a bit of a break before diving into something new.
If you want more excellent Claude Code content, you should follow @addyosmani , @lydiahallie, @_catwu, @ClaudeDevs and of course @bcherny who started it all.
1
1
37
Apollo's chief economist: revenue per worker and margins are up for the Mag 7, flat for the other 493 S&P companies. No productivity bump yet.
That's not "AI doesn't work". Most companies bolted a chatbot onto the same old workflow. The payoff shows up when agents own a job end to end and keep going until it's finished.
apollo.com/wealth/insights-n…
18
NVIDIA just shipped a 4GB open model that runs in 6GB of RAM and handles chat plus tool use. Weights, data and recipes, all open.
Local AI characters that talk and act without a cloud bill are now a weekend project, not a funding round.
teapawt.org/NVIDIAAI/status/211365…
Available now: NVIDIA Nemotron 3.5 Bolt E4B 4GB ⚡️
Our newest small open model is built for local assistants, interactive characters and robots that can carry on a conversation and use tools to get things done.
Watch Reachy Mini chat and move, with Bolt handling conversation and NVIDIA Cosmos handling vision. Both models run locally on Jetson Orin Nano 8GB.
Learn more: research.nvidia.com/labs/nem…
1
1
46
An open 27B model just beat GPT-5.5 and Opus 4.8 on desktop tasks, and nobody touched its weights.
Simular's Agent S4 evolved better interfaces for the model to work through. Qwen3.8-27B now finishes 21.3% of OSWorld V2.
Stop waiting for the next model. Fix the harness.
teapawt.org/SimularAI/status/21136…
Our agent scientist, Agent S4, just got an open 27B model to beat GPT-5.5 and Claude Opus 4.8 on OSWorld V2, with the model's weights left exactly as they were.
With the interfaces S4 evolved, Qwen3.8-27B completes 21.3% of the 108 tasks, the highest binary success of any open-source computer-use agent and 3.8x the same model with fixed interfaces.
It costs $1.10 per task, against $25.50 for GPT-5.5 and $72.40 for Opus 4.8 💸
Full write-up: simular.ai/articles/agent-s4…
2
31
Anthropic just handed Agent Skills to the Agentic AI Foundation, same home as MCP. 70+ platforms already use it.
I run the same skills folder across Claude Code and pi. A SKILL.md that works everywhere is worth more than another model bump.
teapawt.org/ClaudeDevs/status/2113…
We're donating Agent Skills to the Agentic AI Foundation (@AgenticAIFdn), the same foundation we donated MCP to.
Since December 2025, 70+ agent platforms have integrated the standard, including ChatGPT, Cursor, GitHub Copilot, and VS Code.
Here's what that means for you:
2
1
31
Claude just proved 239 fuzz targets across 123 open-source libraries can't crash on any input under 64MB. On the compiled binary.
Fuzzing finds where bugs are. Proofs show where they aren't. That flip is the story.
anthropic.com/research/forma…
12
Most labs train agents to finish a task. Humans& trained Ampersand to help people hit goals over months, and in their multi-month simulations users got more done with it than with Fable.
I run agents that loop until the job's done. The hard part isn't the loop, it's staying in sync with the human.
teapawt.org/humansand/status/21126…
No one changes the world alone
Introducing Ampersand - the first model built for us. Ampersand is a collaborator trained to help people achieve their individual and collective goals. In multi-month simulations, users collaborating with Ampersand accomplished more than with Fable and other frontier models
Available to everyone through our API
18
Anthropic's draft IPO filing, per WSJ: $16.3B revenue in H1, up from $1.1B a year ago. Up to $750B on compute coming.
The line that stuck with me: AWS, Google and Microsoft increasingly see it as a competitor.
When your landlord is also your rival, compute is the real risk factor.
wsj.com/tech/ai/anthropic-re…
29















