Hugging Face is distributing almost 1 exabyte (1 billion gigabytes) of models and datasets to the world every month.
Beautiful to see the live distribution of open intelligence at scale 🌎💪🤗
Announcing the super-adapter for @Alibaba_Qwen Qwen3.8-27B 🔥
We are closing this week with a cool experimental model - a bigger LoRA that merges all harness adapters from this week - with a simple trick: concatenate all LoRA adapters into one.
Result: beats the thinking model, with thinking with much less tool calls under thinking mode: 19-48% fewer with thinking on depending on the harness
Accuracy gains:
SWE-Bench Pro (hard 51):
→ Hermes Agent: 45% → 61%
→ OpenCode: 47% → 67%
→ pi: 74% → 74%
Terminal-Bench 2.1
→ +4 to +6 points under each harness
We also share the full recipe, experimental results, open questions, etc in the longly awaited full recipe blogpost: https://t.co/WxeH2KVhuf
FLUX: In Production
An AI film competition where the studios write the briefs, and you make the film.
We’re partnering with @RTLDE_Corporate , @nurastudios and @SBabelsberg. $25,000+ in prizes, with more to be announced.
Online from 30 Nov to 13 Dec. SF build night Dec 3
Register: https://t.co/0MjypXiMsw
hf-mem v1.0.0 is out, giving you (and your agents) more consistent memory estimates for inference without pulling the weights 🤗
- Weights, KV cache & overhead
- Safetensors, GGUF & ONNX
- HTTP range requests
- Only depends on `httpx2`
More in the thread 🧵
🎉 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! ❤️
OpenShell is joining @CloudNativeFdn 💚
Help shape how AI agents access data, use credentials, and execute code. If you’re building agents, we’d love your input!
Check out the project and contribute 👇
Tokens per second tells you how fast a model writes. It does not tell you whether it wrote the right thing.
At 95% per step, a 10-step agent chain finishes about 60% of the time. New article: what belongs on a model scorecard, and whether quality is even your problem.
Reachy Mini can finally move!
Today We’re launching Reachy Mini "Level Up", a set of new accessories for Reachy Mini: wheels (!), phone apps, and hats that change its personality
obviously the first thing Reachy Mini did was race against Microduck (spoiler: it's faster)
by the way, the team behind microduck is publishing really technical deep dives on how it works
e.g. you can pick the duck up, carry it 1.5m, put it down, and it relocalizes within 2cm. using a 64-pixel depth sensor and leg odometry, all running onboard
great reads by Antoine, data is open on HF too
https://t.co/I4nZYDiMk5
Fast tokens don't tell you whether a model can use tools.
tool-eval-bench v3.0.0 adds a local web UI, shareable results cards and long-context throughput presets. Compare tool-use quality alongside speed.
https://t.co/ERC58VZPPq
@SeraAndroid My colleague came across @agentgateway for for fine-grained control of MCP servers and their tools. As it turns out, it can also be used as a proxy for LLMs. https://t.co/GSxAoFyFdn - pretty impressive. Just started testing. Written in Rust, Apache 2.0, Linux Foundation project.
Kolibri 1 now runs natively in Transformers 5.20!
We are still overwhelmed by all the community feedback.
You can now spin up our Kolibri models in just a few lines of code using 🤗transformers.
Special thanks go to Luca Motz (@lucamotz3141) and Harry Mellor (@hmellor_) for pushing this.
Just watched the @netflix Instadocs about the @OpenAI cyberattack on @huggingface. Super disappointing, as they went for maximum anthropomorphism of the technology and full doomer mode.
I spent more than two hours with them trying to correct all their mistakes and ground this in reality, explaining that agents are technological systems, not aliens. But it looks like the drive to scare people for maximum views was too strong.
Also, they didn’t mention the lack of containment and monitoring of the agents, how closed-model safeguards blocked our defensive work, or how we defended ourselves with AI using an open model. And that we still lack a lot of information and transparency, including full agent traces and that companies need to remain accountable for what their agents do.
It ends up being a fun sci-fi movie though!
It feels pretty fitting for @JensenHuang & @nvidia to end their European tour with Microducks on stage.
Microduck was designed and engineered in Europe (more precisely Bordeaux, France) by the @pollenrobotics and @huggingface teams. With American capital and ambition + Chinese manufacturing genius, it became the most popular robot of 2026 with over 15,000 sold all over the world!
I believe Europe has more opportunity to shine in AI than in any previous tech wave. We intend to prove it with @huggingface@nvidia and the whole open-source AI community in the coming years!
Let's go European AI 🇪🇺🇪🇺🇪🇺