fable 5 is the real deal. of course the benchmarks are great but *qualitatively* it’s a real pleasure to use and the biggest step up in model performance since opus 4.5
it was also one the first models i had the pleasure of working on at Anthropic, so it’s amazing to see it go live and have people try it out.
🧡
Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use.
Its capabilities exceed those of any model we’ve ever made generally available.
i recently did a talk on this at @swyx 's ai engineer conference - i think i did a pretty bad job at delivering it (jetlag rip)
the ideas from it were pretty good tho so just surfacing the post in case anyone is curious - (core work done by @rgb_prithvi)
https://t.co/HckWJmpzTc
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.
When do you reach for other models instead of Claude? What can we do better? Hit me with all of your frustrations. dms open.
If you can give me detail (e.g. specifics/transcipts) - it'll help a lot in finding out exactly what we need to do to improve the next model
there’s a lot of terrible takes out there but this isn’t one of them. captures a lot of feel for the London startup ecosystem that takes years to develop - nice work @Zainmbrk
https://t.co/otbTNXI45f
Automating AI research is the next major step in AI
We let Claude Code (Opus 4.7) and Codex (GPT 5.5) run autonomously on the nanoGPT speedrun optimizer track using our idle compute. ~10k runs, ~14k H200 hours
Opus now holds the record at 2930 steps vs the 2990 human baseline
We’ve agreed to a partnership with @SpaceX that will substantially increase our compute capacity.
This, along with our other recent compute deals, means that we’ve been able to increase our usage limits for Claude Code and the Claude API.
a lot of startups should be more ambitious what what they think an intelligence unconstrained version of their products look like
beyond automating what a human doing the task end to end would do even
lots of alpha here