SOTA Mathematical Reasoning on IMO-AnswerBench and PapersWithCode

84.5Accuracy

DeepSeek-V3.2

Updated 12d ago

Evaluation Results

Method
DeepSeek-V3.2 2025.12	84.5	45
Gemini-3.0 2025.12	83.3	18
Gemini 3 Pro 2026.04	83.2	-
Gemini-3.0 2026.02	83.1	18,000
Kimi K2.5 2026.02	81.8	36,000
Kimi K2 2026.02	78.6	37,000
Kimi-K2 2025.12	78.6	37
DeepSeek-V3.2 2026.02	78.3	27,000
DeepSeek-V3.2 2025.12	78.3	27
QED-Nano (+ RSA test-time scaffold) 2026.04	76.5	-
GPT-5 2025.12	76	31
DeepSeek-Math-V2 2026.04	75.8	-
Qwen3-235B-A22B-Thinking-2507 2026.04	70.5	-
GPT-OSS-120B 2026.04	70.5	-
QED-Nano 2026.04	67.5	-
Qwen3-30B-A3B-Thinking-2507 2026.04	67	-
GPT-OSS-20B 2026.04	61.5	-
QED-Nano (SFT initialization only) 2026.04	57.5	-
Qwen3-4B-Thinking-2507 2026.04	55.8	-
Nomos-1 2026.04	49	-