Charlie Marsh
@charliermarsh
@OpenAI. Building Ruff, uv, ty, and other high-performance Python tools with the @astral_sh team.
998 Following 53.8K Followers
Caring is the moat, you should feel visceral pain when things are bad or broken and stop at nothing to fix them
An update here: we’re upgrading all Codex for OSS grants to Pro 200!
There will still be the 10,000 grants in this round, now at the higher tier. (If you've already received a grant, the upgrade should be reflected automatically.)
Applications are still open :)
Show more
We're starting to see some agentic Rust forks (or alternative implementations) focused on iteration speed.
More-so than the specific projects, I'm interested in the ideas being explored (and how they could evolve upstream), so wanted to collate a few here...
Show more
Getting people to care about what you built remains arguably the hardest part of building software
@sharifshameem I looked and my acceptance rate is is 32.0%, I might be cooked
These are truly so, so good
My conversation with Head of ChatGPT and Codex, @thsottiaux
We discuss:
🔸 Why the model picker is likely going away
🔸 Why loops and graphs are a passing phase
🔸 Why most actions on the internet will soon be taken by agents
🔸 OpenAI’s approach to AI safety
Listen now 👇
Show more
If Rust fades, my belief is that it will be for this reason
Rust no longer the defacto best language for agents
Compile times are too long, I'm always finding myself bottlenecked by time not intelligence
And the agents have gotten really at doing manual memory management now
Show more
gpt-6.1 sol has saturated the FrontierMath Tier 4 benchmark that was considered exceptionally difficult even for research-level math just last year
that's only 10 months of progress
also Astra solved 5 erdős problems, and 2 of them under budget
Show more
This is crazy. My Dot proactively messaged me to point out that a bunch of our CI jobs would start failing on Monday due to GitHub's macos-14 runner retirement.
I jinxed it, delayed a ~week, rc3 coming next instead
Python 3.15.0 stable ships today
You are CPython 3.15, a general-purpose Python interpreter trained on the entire standard library.
Python 3.15.0 stable ships today
GPT-6.1 Sol is our most demanded model pretty much ever both across both the API and subscriptions.
Within ChatGPT & Codex, we were under heavy load, but have brough more capacity online and the speed should get much better in the coming hours, reaching almost twice the speed compared to what we served yesterday.
Show more
GPT-6.1 Sol from @OpenAI on ARC-AGI (Verified):
- ARC-AGI-3: 52.7%, $7.6K (standard harness), 96.4%, $4.4K (provider adapter harness)
- ARC-AGI-2: 94.2%, $0.25/task
- ARC-AGI-1: 98.5%, $0.06/task
Its 96.4% on v3 was comparable to GPT-6 Astra's 99.9% but at a 77% lower cost.
Show more
When I see a crate with extremely judicious allocation discipline
RT @thenanyu: It’s interesting to contrast, what different companies take for granted in their internal cultures
I’ve only been at OpenAI…
we found opus 5.5 to be 2X as efficient as gpt-6-sol in our internal evals so i was not expecting 6.1-sol to improve on that much
but to see it 3X the efficiency of opus 5.5 and 2.5X against even sonnet 5.5 is actually 🤯
this is an insane point release
Show more
I also have a little announcement for DevDay. I asked Tibo to include it in the presentation but he didn't read my slack DM 😭
My first onboarding project is merged. I fixed this issue once and forever! I reduced amount of disk writes by 50% and the size of stored files by 4x
Show more
GPT-6.1-Sol gets #2# on eyebench-v3, stealing Opus-5.5s short-lived spot!
It's near-astra performance for ~3.8x cheaper, and ~8x cheaper than Opus-5.5 while scoring better.
