Modern problems require modern solutions

GPT-5.1 is out! It's a nice upgrade. I particularly like the improvements in instruction following, and the adaptive thinking. The intelligence and style improvements are good too.

I've been testing GPT-5.1 for a few days. My quick notes: - creative writing style is a LOT better - it's much faster than GPT-5 (with similar intelligence) for most prompts - the personality is WAY better (but can still sometimes be annoying) - it's great in Codex!

GPT-5 on Sudoku-Bench 🧩 Since releasing Sudoku-Bench in May 2025, when no LLM could solve a classic 9x9 puzzle, we've been evaluating the latest generation of models. GPT-5 now leads our leaderboard with 33% puzzles solved--approximately 2x the previous leader--and is the first

I finally reached human-level performance (85%) on ARC-AGI v1 for under $10k and within 12 hours. I use the same multi-agent collaboration with evolutionary test-time compute, now powered by GPT-5 pro with lower parallelism.

congrats to llama 3 large for winning the LLM trading contest by not participating

met the great genius @GMJuditPolgar yesterday

Still, what an honor: to play, to learn, and to help bring @GMJuditPolgar to @OpenAI. An experience I’ll be chasing for a long time.

I had one final amusing moment. After Judit played the crushing 35. e5, I found 35… g6, which appears convincing and in a bullet game (my specialty), might have actually saved the game. But of course she found the (only) refutation in seconds 😕

These lines are not too complicated, but under the pressure of playing Judit, and knowing that my game was broadcast to hundreds of my colleagues, I crumbled and ended up losing all my queenside pawns in the next 5-10 moves…

Surprisingly, when I later checked with the engine, it turned out that everything just works for Black. 19 Nc4 d5 is a sound positional pawn sacrifice, and after 19 Ra5, Bd8 is the top move, though the core idea is 20… a5!, neutralizing the rooks rather than trading bishops.

The critical position in my game came after Judit played Ra5 on move 19, preparing to double rooks while eyeing the key d5 square. I had missed this idea and instead focused only on 19. Nc4 d5!?. After Ra5 I panicked and embarked on a wild idea to trade bishops with Be7-d8-b6.

Playing Judit was surreal. She’s the strongest opponent I’ve ever faced, and it showed immediately, in her precision, tactical prowess and strategical finesse. Just seeing her across the board was enough to lose all self-confidence.

Then came the highlight of the event, the simultaneous match. 8 OpenAI researchers vs. Judit. Final score: 0.5–7.5. We joked that we exceeded expectations 😅

Judit shared stories about her work with Ernő Rubik and the Judit Polgar Chess Foundation, a reminder that learning, play, and curiosity belong to everyone. 🔗 https://t.co/QweP7rHEAm

The event started with a fireside with @markchen90. They talked about chess before and after engines, how AI can sharpen or stifle creativity, and the beautiful ways human imagination and the classic game of chess persists beyond engines.

Three weeks after joining OpenAI, I sent Judit a cold email: “it would mean the world to me if you could make it. Both your chess style and life story are a source of continued inspiration, and I'm sure many of my colleagues (chess players or not) will love to hear from you.”

Last week, we hosted Grandmaster @GMJuditPolgar, a legend and the strongest woman in chess history at @OpenAI for a conversation on intuition, creativity, and strategy. We think about these ideas every day. Chess is a reminder that reasoning and imagination go hand in hand.

This is absolutely insane! 2800 with two pieces down...

One of my bucket list items were checked off today: we hosted @GMJuditPolgar at the @MILibrary chess club!