Liquid's LFM2.5-8B-A1B smashed OpenAI's gpt-oss-20b on tool calling
We ran both locally on a MacBook Pro M5 Max, 64GB, and gave each the same trip-planning request that only completes if the model fires all 7 tool calls - weather for 3 cities, two currency conversions, an email and a reminder
Outputs:
LFM2.5-8B-A1B: 4.8 GB RAM usage, 7/7 tool-calls, 266 tok/s, 6.9s
OpenAI gpt-oss-20b: 11 GB RAM usage, 3/7 tool-calls, 146 tok/s, 15.0s
The 8B used less than half the RAM and still fired all 7 calls, while the 20B silently dropped more than half of its own. It also ran ~2x faster, wrapping the full agentic request in 6.9s against 15s. That's what 38T training tokens buy: a 1B-active MoE that nails the agentic tool calls a model 2.5x its active size keeps dropping
Today is my first day at @OpenAI! I'm joining the Codex team. I'll be splitting my time between building the Codex app and working with developers/entrepreneurs to help them make the most of Codex.
As a @YCombinator alum, I'm really looking forward to working with early-stage startups and couldn't be more energized about OpenAI's $2M investment in new YC startups.
Despite OpenAI being the largest company I've been part of, it certainly feels like @sama, @gdb and the whole Codex team are in founder mode. I've never seen another startup move as fast as this team does. I'm truly honored to be joining such a brilliant team.
As always, it's all about the people... so I'd like to thank @romainhuet, @dkundel, @coreyching, @ajambrosino, @embirico, @thsottiaux, and the rest of the Codex team for their trust.
And thanks to @pankaj, @gilad, and @willhorn for believing in me, bringing me back to Silicon Valley after almost ten years, and giving me the opportunity to learn about AI.
Despite public statements from President Trump and Tulsi Gabbard citing her husband’s diagnosis with a rare form of bone cancer as the reason for her departure, Reuters is reporting that this was not a choice by Gabbard, and that the White House forced her to resign from her position as Director of National Intelligence.
Today, we share a breakthrough on the planar unit distance problem, a famous open question first posed by Paul Erdős in 1946.
For nearly 80 years, mathematicians believed the best possible solutions looked roughly like square grids.
An OpenAI model has now disproved that belief, discovering an entirely new family of constructions that performs better.
This marks the first time AI has autonomously solved a prominent open problem central to a field of mathematics.
Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.
We ran an LLM onboard a satellite to describe Earth! The response comes from @liquidai LFM2 - marking the successful commissioning of our mini orbital server.
Read the full story: https://t.co/AoaYXeHZ5b
Run your own software in space: https://t.co/Rfbyyifbyv
No Trump judges, no voter ID SAVE Act, 50 bills already passed by the house in limbo.
Pardon my language, but this guy is a worthless piece of shit. He’s worse than a grifter Democrat. Fetterman has done more for the Republican platform than this weasel. Treacherous rat.
I was the first Governor to endorse Obama for President back in 2007. I was there when he announced in Springfield Illinois. I have known him since 1995. Do I think it likely he ordered a phony intelligence assessment in December of 2016 to destroy Trump & his new administration? ABSOLUTELY!
Demis Hassabis: "In the near future, one person who knows AI will outperform an entire startup team"
I've watched hundreds of AI talks, this 60-minute Cambridge lecture is the one I wish I had seen a year ago
this is the Nobel Prize winner in Chemistry, CEO of Google DeepMind and the guy who made AI solve biology
here's the part I can't stop thinking about:
> the AI you're using today is the dumbest it will ever be
> in 5 years the gap between people using AI and people who aren't will be impossible to hide
> companies will run on 10 people doing what 200 used to do
> the ones who get there first won't be the smartest, they'll be the ones who started right now
right now the average person opens Claude, types something, gets an answer, closes the tab
they think they're using AI, but they're using maybe 10% of it
I turned his lecture into 18 steps to actually use Claude the way it was designed, copy-paste prompts included
full guide in the post below.