BlogLLM Benchmark Scores as Published, Side by Side

MathArena Apex: Results as Published

Every result the catalog holds for MathArena Apex: 19 models from 10 developers, read from 4 pages. Each number is as printed by the company named beside it; nothing is ranked or rescored. Models are listed by developer, newest first.

Area Science and MathMaintainer SRI Lab at ETH Zurich and INSAITIntroduced 2025Data version 2026103001, as of 2026-10-30

MathArena Apex is a small set of final-answer competition mathematics problems that frontier models could not solve when it was built. The launch set (18 August 2025) has 12 problems filtered from nearly 100 competitions, mostly from 2025: six from final-answer competitions and six converted from proof-based ones such as the IMO and team selection tests. A candidate problem was dropped if any of four frontier-model attempts answered it. The score is final-answer accuracy averaged over problems and repeated attempts, and the maintainers add problems over time.

MathArena Apex Shortlist

ModelValuePrinted bySetup as printed
GPT-5.4OpenAI, released 2026-03-0578.1%DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15Effort: xhigh
Claude Opus 4.6Anthropic, released 2026-02-0585.9%DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15Effort: max
Gemini 3.1 Pro PreviewGoogle, released 2026-02-1989.1%DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15Effort: high
DeepSeek-V4-Flash (Preview)DeepSeek, released 2026-04-249.3%DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15; printed on 1 more pageEffort: non-think
72.1%DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15Effort: high
85.7%DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15Effort: max
DeepSeek-V4-Pro (Preview)DeepSeek, released 2026-04-2490.2%DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15; printed on 1 more pageEffort: max
9.2%DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15Effort: non-think
85.5%DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15Effort: high
DeepSeek-V4-Flash-0731DeepSeek88.7%NVIDIA, NVIDIA Nemotron 3.5 Super VL 120B-A12B model cardchecked 2026-10-19-
Qwen3.5-122B-A10BAlibaba, released 2026-02-2454.7%NVIDIA, NVIDIA Nemotron 3.5 Super VL 120B-A12B model cardchecked 2026-10-19-
Qwen3.5-397B-A17BAlibaba, released 2026-02-1661.4%NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15Tools: none
60.4%NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15Tools: with tools
Qwen3.8-Flash-NextAlibaba91.8%NVIDIA, NVIDIA Nemotron 3.5 Super VL 120B-A12B model cardchecked 2026-10-19-
Kimi K2.6Moonshot AI, released 2026-04-2075.5%DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15; printed on 1 more pageEffort: thinking
GLM-5.3-FlashZ.ai, released 2026-08-2677.1%NVIDIA, NVIDIA Nemotron 3.5 Super VL 120B-A12B model cardchecked 2026-10-19-
GLM-5.1Z.ai, released 2026-04-0772.4%DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15; printed on 1 more pageEffort: thinking
MiniMax M2.7MiniMax, released 2026-03-1851.9%NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15Tools: with tools
28.9%NVIDIA, NVIDIA Nemotron 3 Ultra 550B-A55B model cardchecked 2026-10-15Tools: none
Nemotron 3.5 Super VL 120B-A12BNVIDIA, released 2026-10-1588.36%NVIDIA, NVIDIA Nemotron 3.5 Super VL 120B-A12B model cardchecked 2026-10-19-
Nemotron 3 UltraNVIDIA, released 2026-06-0475%NVIDIA, NVIDIA Nemotron 3.5 Super VL 120B-A12B model cardchecked 2026-10-19; printed on 1 more page-
Inkling-SmallThinking Machines71.5%NVIDIA, NVIDIA Nemotron 3.5 Super VL 120B-A12B model cardchecked 2026-10-19-

MathArena Apex

ModelValuePrinted bySetup as printed
GPT-5.4OpenAI, released 2026-03-0554.1%DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15Effort: xhigh
Claude Opus 4.6Anthropic, released 2026-02-0534.5%DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15Effort: max
Gemini 3.1 Pro PreviewGoogle, released 2026-02-1960.9%DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15Effort: high
DeepSeek-V4.1-FlashDeepSeek, released 2026-09-1065.6%DeepSeek, DeepSeek-V4.1-Flash model cardchecked 2026-10-15Effort: max
DeepSeek-V4-Flash (Preview)DeepSeek, released 2026-04-241%DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15Effort: non-think
19.1%DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15Effort: high
33%DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15Effort: max
DeepSeek-V4-Pro (Preview)DeepSeek, released 2026-04-2438.3%DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15Effort: max
27.4%DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15Effort: high
0.4%DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15Effort: non-think
DeepSeek-V4-Flash-0731DeepSeek58.6%DeepSeek, DeepSeek-V4.1-Flash model cardchecked 2026-10-15Effort: max
DeepSeek-V4-Pro-0813DeepSeek65.3%DeepSeek, DeepSeek-V4.1-Flash model cardchecked 2026-10-15Effort: max
Kimi K3Moonshot AI, released 2026-07-1665.6%DeepSeek, DeepSeek-V4.1-Flash model cardchecked 2026-10-15-
Kimi K2.6Moonshot AI, released 2026-04-2024%DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15Effort: thinking
GLM-5.1Z.ai, released 2026-04-0711.5%DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15Effort: thinking

Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.