Highlights from this release: now tool calls run in parallel in ToolCallingAgent, meaning that task execution can get Nx faster!
@aymericroucher
-
LLMs now understand physics, disproving skeptics
By
–
One year ago, LLM skeptics were saying "image models do not understand physics and never will", and they would cite as examples messed up images of gymnasts.
— m_ric (@AymericRoucher) 24 juin 2025
Now it appears LLMs can represent complex dynamics accurately. So they must 'understand' physics to some extent.
I think… https://t.co/tSxBkLxwRpOne year ago, LLM skeptics were saying "image models do not understand physics and never will", and they would cite as examples messed up images of gymnasts. Now it appears LLMs can represent complex dynamics accurately. So they must 'understand' physics to some extent. I think
-
US Datacenter Power Demand: 88GW deficit by 2028, nuclear comparison
By
–
I really recommend watching The Geopolitics of AI Infrastructure by @dylan522p > Following the current trend and datacenter power demand, the US will be lacking ~88GW by 2028. For reference, 1 nuclear reactor provides ~1GW. There's a long way to go!
-
Smolagents surpasses 20k stars on GitHub celebration
By
–
Almost didn't take time to celebrate : Smolagents just wooshed past 20k stars on Github! Congrats to the team, especially @avillanovamoral who spends lots of time improving the library. And thanks a lot to our community members for pushing is forward! It's still time to
-
Agents Hack next Sunday: 10k€ prizes with Hugging Face, AnthropicAI, Accel
By
–
Reminder: our Agents Hack is next Sunday ! It's shaping up to be the biggest ever hackathon on Agents in Paris, with over 10k€ in prizes, including large cash prizes for the top 3 teams. And now, @huggingface and @AnthropicAI are joined by @Accel : this will bring a bit of
-
12B model could memorize entire English Wikipedia given bit density
By
–
English Wikipedia is 29.4B characters, with ~1.5 bits of information per character. So that means, at 3.6 bits per parameter, a 12B model could memorize all english Wikipedia, it seems wild!
-
ScreenSuite: Comprehensive Evaluation Suite for GUI Agents
By
–
Today we release ScreenSuite, the most comprehensive evaluation suite for GUI agents (aka Computer Use agents). We packed 13 benchmarks, and 3 different environments, to evaluate the full range of agentic capabilities for vision models. And it turns out, @Alibaba_Qwen models are
-
Qwen unveils new high-performance multilingual embedding standard
By
–
Qwen silently dropped the new standard for embeddings on the Hub!
– 0.6B, 4B and 8B versions (probably would use only the 0.6B)
– 32k context length – 100 languages – SOTA on MTEB, but like real SOTA, with 10 points margin on the second bests https://
x.com/tomaarsen/stat
/tomaarsen/status/1930579927020994694
… -
SmolVLA report introduces Async inference to boost robot actions
By
–
If you didn't yet, you should read the technical report for SmolVLA, published yesterday by the @huggingface robotics team!
Amongst other ideas, it introduces "Async inference" to boost their robot actions. Robots have a problem: performing the actions takes time (Unlike -
Agents Hackathon with Hugging Face and Anthropic in Paris
By
–
Save the date! => Agents Hack w/ Hugging Face + Anthropic June 15th, Paris We're teaming up with Anthropic and Unaite to organize the largest agents hackathon, with 10k in prizes! It's going to be a full day of hacking on June 15th, at Hugging Face's Paris office. Spots