OpenAI MedChemBench - Scientific Explanation: leaderboard

OpenAI's MedChemBench (20 November 2026): 99 problems built from 117 medicinal chemistry papers and patents, where a model gets a drug program's target, goals, compounds and results so far and must reconstruct structures, explain findings, pick compounds and predict properties. One attempt per response, no tools, ten responses per problem. This board is the explanation component only: whether the model's reasoning, such as why one compound was chosen over another, is supported by the evidence supplied. Rows are model and reasoning-effort pairs for eleven OpenAI models; GPT-5.2, GPT-5.4 and GPT-5.5 come from the paper's tables. Share of rubric credit (%); higher is better.

Metric: Average credit (%). Source: openai.com. Saturation forecast: Around February 2027. 11 models tracked. Updated 21 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1GPT-6 AstraMax91.74#10
2GPT-6.1 SolMax90.02#15
3GPT-5.6 SolMax76.41#37
4GPT-6 SolMax75.3#34
5GPT-6.1 TerraMax75.25#25
6GPT-5.5High66.99#54
7GPT-5.6 TerraMax61.3#71
8GPT-5.6 LunaMax60.63#131
9GPT-5.2xHigh59.51#185
10GPT-5.4xHigh58.52#135
11GPT-6 LunaMax55.64#105

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=openai-medchembench-scientific-explanation · How It Works · Data refreshed daily, snapshot 2026-11-21.