Thank you,
Shreyas Pimpalgaonkar
We built this to say thank you. Everything on it is public and cited, and none of it came from anywhere you did not put it yourself.
He works on the unglamorous half of frontier AI: the environments a model is actually evaluated and trained in.
Shreyas Pimpalgaonkar works at Bespoke Labs on post-training and reinforcement-learning environments, which is how he describes his own focus. Bespoke Labs builds environment infrastructure for AI agents in production, spanning company-scale simulations with real codebases, agent evaluation and optimisation, and production-grade reinforcement learning and benchmarking.
He is a co-author of Terminal-Bench, submitted to arXiv in January 2026, a benchmark of 89 hard, realistic command-line tasks, each with its own environment, a human-written solution and comprehensive verification tests, across 16 categories from software engineering and security to scientific computing and debugging. He is also a co-author of Bespoke-MiniChart-7B, an open vision-language model for chart understanding that reached state-of-the-art chart question-answering results among 7-billion-parameter models.
Speaking at the AI Summit at Stanford on Keynote: Beyond Scale: The Data Quality Problem in Frontier AI.
This page exists to say thank you for showing up and doing the work in public. Every claim on it is cited below and checkable. The fuller picture our research put together is not published here, and will not be. It is yours the moment you claim it.
The record
Every line below is cited to a primary source you can open.
Works on post-training and RL environments at Bespoke Labs
Shreyas PimpalgaonkarCo-author of Terminal-Bench, an 89-task benchmark for agents in command-line environments, arXiv January 2026
arXivCo-author of Bespoke-MiniChart-7B, an open vision-language model for chart understanding
Bespoke Labs
@shreyas-pimpalgaonkar
This name is held for you.
This name is held for you, not by us. Claim it and the fuller model we researched becomes yours, on your own storage, for you to correct or delete.
There is more we found than is shown here. It is not published and it will not be. It is yours when you claim it.
Sign in and it is yours.
Shreyas, one tap is all this takes. Use an account you already have and @shreyas-pimpalgaonkar is yours. Nothing new to sign up for, and no password to invent.
Signing in only proves the name is yours. Nothing is posted anywhere, nothing is shared, and it stays free.
Your 🤫 ID in Apple Wallet
Both cards, your spaceID and your 🤫 ID, are designed to live in Apple Wallet and Google Wallet so a check-in is a tap rather than a form.
Not yet issuable. A pass has to be signed with an Apple Pass Type ID certificate, and we do not hold one yet. The pass itself is built and waiting; we would rather say so than show you a button that quietly fails.
Not happy this exists?
Ask us to take this down and it goes, the same day, with nothing kept. Tell us and it is gone.
Where this came from
- Role and session read from the event programme, captured 31 July 2026, via AI Summit at Stanford
- Current affiliation, focus area and publication list, self-published, via Shreyas Pimpalgaonkar
- Co-authorship, title, scope and date of Terminal-Bench, via arXiv