Comment by MostlyStable

That's a good point. The coding specific question (which still uses the arena.ai leaderboard, just the coding version instead of text version), has Anthropic at 66% at Anthropic/OAI at just shy of 80%. That's probably more representative even if Arena.ai isn't my favorite benchmarking site.

Actually....looking in more depth I may have to upgrade to something stronger than "not my favorite benchmarking site". The first OAI model on there is GPT 5.4 at 31....that's not really credible.

https://polymarket.us/event/code1-2026-12-31