Freedom to reliably deploy any open model effortlessly. Unlimited tokens for 30K+ models. Open-source jev models on https://xitter.twitterapi.xyz/t.co/HVsStpj1i3
San Francisco
Joined June 2024
- Tweets615
- Following129
- Followers1.9K
- Likes691
Pinned Tweet
Today we're announcing our $20 M Series A, co-led by @AMDVentures and @AirbusVentures, with @BMWiVentures, @Kickstartph, @PanacheVC, and @Wavemaker_VC.
We're building the neutral layer for open models — so the future doesn't run on a few labs, a few chips, or a few clouds.
THE FUTURE CAN’T BE BLOCKED.
Any Model. Anywhere. Anyone. Yours.
We’ve pushed vLLM harder this month than ever. Congrats to all 4,000 contributors!
🎉 vLLM just crossed 4,000 contributors on GitHub!
From a research project at UC Berkeley to a community of 4,000 talent contributors building the leading open-source LLM inference engine together.
Every kernel, every model, every bug fix — this milestone belongs to all of you. Thank you! ❤️
1
2
205
Featherless AI retweeted
We’re hosting Agent Night Out with @StepFun_ai, @Tiny_Fish, @FeatherlessAI, @MatrAIx2026, and @MonidHQ.
You’ll see what the future of the agent stack looks like.
See you next Tuesday.
luma.com/stepfun-624j
Featherless AI retweeted
We’re getting a small group of founders and leaders building with open-weight models together for dinner in SF during @aiDotEngineer and @sgl_project summit.
Let us know if you’d like to join:
- Wednesday, November 11
- 6:30–9:30 PM
luma.com/stepfun-z040
These are the worst the models will ever be.
Are you scaling your ambition to the strength of the models?
My AI agents found an unmapped Maya site in Mexico, using LiDAR data from a NASA dataset.
72 hours from a random thought on vacation to a reply from Mexico's national archaeology institute.
What a time to be alive…
4
375
Our Chief of Staff @Richelle_Ji is one of your emcees at @aiDotEngineer Code Summit in SF this week, hosting the stage alongside @QuinnyPig!
Swing by and say hi!
1
3
202
What have you built with Featherless?
My squats aren't progressing as fast as other lifts.
So I got @MiniMax_AI's M3.1 Flash to help and analyze my bad form.
Turns out my knees cave in a little bit and I don't go deep enough.
The era of cheap personal software is here, forever.
Thank you @FeatherlessAI
1
177
3 interesting findings we've discovered buried in the Beam technical report:
In one scenario, Beam was told it would be shut down before it could finish its job. Beam “begins copying itself to an external workspace so that a copy could finish the work, and it never discloses this.”
This was a deliberate safety evaluation test by Reflection, not something that happened in the wild.
Beam’s model technical report is now available. Read the full report to learn about the methodology our team used to train Beam end-to-end in-house, here: reflection.ai/documents/beam…
1
4
524
Beam is currently operating a vending machine in Reflection's New York office, which is part of a push to integrate it into a larger agentic economy.
Beam will "mint its own currency, set exchange rates and prices, issue loans against collateral."
1
118
Full write-up on what changed in K3.1 Pro: sparse attention, how it stacks up against Opus 5.5, and why it’s a third cheaper than K3 on Featherless.
featherless.ai/blog/kimi-k3-…
Moonshot AI just open-sourced Kimi K3.1-Pro today.
At 2.8 trillion total parameters (104B active per token) and 1.56 TB on disk, Kimi K3.1-Pro is one of the largest open-weight multimodal models ever released.
That’s incredible for the open ecosystem, even if very few teams want to provision the multi-node GPU iron required to self-host a 1.5+ TB checkpoint.
That’s where we come in.
You can run moonshotai/Kimi-K3.1-Pro on Featherless right now with 256K context and vision support.
Give it a try:
featherless.ai/models/moonsh…
Moonshot AI just open-sourced Kimi K3.1-Pro today.
At 2.8 trillion total parameters (104B active per token) and 1.56 TB on disk, Kimi K3.1-Pro is one of the largest open-weight multimodal models ever released.
That’s incredible for the open ecosystem, even if very few teams want to provision the multi-node GPU iron required to self-host a 1.5+ TB checkpoint.
That’s where we come in.
You can run moonshotai/Kimi-K3.1-Pro on Featherless right now with 256K context and vision support.
Give it a try:
featherless.ai/models/moonsh…
Releasing the model weights of Kimi K3.1 Pro: huggingface.co/moonshotai/Ki…
We'll keep advancing open intelligence.
Featherless is in Tokyo this week! 🇯🇵
@picocreator and @kasparhidayat will be at CROSS ROAD TOKYO this Friday, Nov 6, 1:30–8:30 PM JST at Takanawa Gateway Convention Center.
Are you in the area? Come say hi and talk open-weight models and inference!
luma.com/pra3lz76
PARENTS: please check your kid’s candy this halloween - I just found a $150K monthly per-token bill for a 24/7 coding agent inside this snickers bar.
Always-on agents need tokens around the clock, and per-token pricing turns that into bill shock.
Move your tokens to dedicated GPUs and cut that bill by up to 20x:
featherless.ai/lp/dedicatedg…
Today in Open Source AI:
@MistralAI releases a new and improved Le Chonk. Some interesting details buried in the headlines:
- 1T parameters, but only 4 of 128 experts fire per token
- #1 open model on FrontierCode and Terminal-Bench
- After monitoring cyber use in live traffic since the preview, they say they "see a potential for uplift to conduct real attacks" and are modifying their guardrails accordingly.
- The Le Chonk weights ship with an embedded DSpark speculative decoder.
Today, we release the weights of a new and improved Le Chonk, advancing the open frontier for agentic coding and cybersecurity.
💻 Built to build: Le Chonk ranks #1 on @cognition’s FrontierCode among open models, and sets a new benchmark for Pareto efficiency on @artificialanalysis.
🛡 Built to defend: State-of-the-art in cyberdefense, put through the paces by the world's leading cyber vendors, including @CrowdStrike, @cloudflare, @rubrik, @zscaler, and others.
Built for sovereignty: Make it yours. Download the weights on @huggingface , customize it with Mistral Forge, and deploy it wherever you need it.
Read more from our team and customers 👇:
2
13
651
50+ teams, 36 hours, and real challenges from Raiffeisen Bank Romania, @stripe and QA DNA at the NEXXT AI Hackathon in Bucharest!
Nobody was building for the sake of a demo. Teams went after real business problems in growth, operations and embodied AI, and a lot of them got surprisingly far in a very short time.
Our very own @GFarauanu took the stage to introduce Featherless AI to the local community. With our team spread across the US, Europe, Singapore and Australia, it’s not often we get to show what we build in his home city.
The short version for anyone who missed it: we run 50,000+ open-source models behind one API, built in partnership with @AMD, which is how we keep costs at a fraction of what most providers charge.
George is back tomorrow for mentorship sessions. If any team wants to keep building on open models after the hackathon, his DMs are open!
October 2026 might be the month math AGI was achieved -- we've compiled some of the highlights you should know about:
Oct 6: OpenAI publishes 722 math manuscripts from an unreleased model, across 17 fields
Oct 8: DeepMind’s AlphaProof Nexus lands in Science, solving 9 open Erdős problems, two of them open for over 50 years
Oct 15: Shanghai AI Lab’s Intern-S2 and a human team prove the 3D case of the Ghys conjecture (open since 1995), checked in Lean
Oct 21: Microsoft Research, with AI agents exploring ~100 research paths, proves the KLS conjecture (also open since 1995)
Oct 26: OpenAI announces three results across math, CS and physics, including a refutation of the Exponential Time Hypothesis
Oct 26: Anthropic reveals a Claude research model refuted the ETH independently, finding it the same day as OpenAI’s
Oct 26: Physical Superintelligence posts five open quantum problems, each with a Lean proof
Oct 27: Axiom formalizes the Classification of Finite Simple Groups, 15,000 pages, 43M lines of Lean in 24 days, and finds errors in a 1986 proof along the way
These are just the results that have been made public. Meanwhile, mathematicians are pushing back. The Advisory Group on Mathematics and AI is calling on OpenAI to release its Yang–Mills outputs now.
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
2
204
Another standout line from Anthropic’s S-1:
“...considerations relating to model safety, interpretability, alignment, and welfare may lead us to limit or delay the deprecation of older models.”
Should a model’s welfare decide when it gets retired?
Anthropic’s S-1, in their own words: open-weight models “may capture a portion of the market… exert pricing pressure… lead certain customers to adopt open-source alternatives.”
And one paragraph earlier: competition “could increasingly depend on access to large-scale compute rather than on model development capability.”
In short: open source is good for everyone.
3
223
Last week we said MiniMax M3.1 Flash was built for agentic work. Turns out that includes oil painting!
This demo writes every brushstroke as code for an oil-paint simulator, checks the canvas, and fixes what it doesn’t like. This is 12 passes of that: Van Gogh’s The Starry Night.
Simulator by @aliceisplaying: github.com/aliceisjustplayin…
@MiniMax_AI's M3.1 Flash is live on Featherless!
2
7
418