BlogLLM Benchmark Scores as Published, Side by Side
MATH: Results as Published
Every result the catalog holds for MATH: 10 models from 4 developers, read from 3 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 Dan Hendrycks and co-authors (UC Berkeley)Introduced 2021Data version 2026103001, as of 2026-10-30
MATH tests competition mathematics with 12,500 problems from high-school competitions, each with a step-by-step solution. The score is the share of problems with the correct final answer.
MATH
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| Llama 4 Maverick (pre-trained)Meta, released 2025-04-05 | 61.2% | Meta, Meta Llama 4 model cardchecked 2026-10-15 | - |
| Llama 4 Scout (pre-trained)Meta, released 2025-04-05 | 50.3% | Meta, Meta Llama 4 model cardchecked 2026-10-15 | - |
| Llama 3.1 405B (pre-trained)Meta, released 2024-07-23 | 53.5% | Meta, Meta Llama 4 model cardchecked 2026-10-15 | - |
| Llama 3.1 70B (pre-trained)Meta, released 2024-07-23 | 41.6% | Meta, Meta Llama 4 model cardchecked 2026-10-15 | - |
| DeepSeek-V4-Flash-BaseDeepSeek, released 2026-04-24 | 57.4% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15; printed on 1 more page | - |
| DeepSeek-V4-Pro-BaseDeepSeek, released 2026-04-24 | 64.5% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15; printed on 1 more page | - |
| DeepSeek-V3.2-BaseDeepSeek, released 2025-12-01 | 60.5% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | - |
| Kimi K2 BaseMoonshot AI, released 2025-07-11 | 70.2% | Xiaomi, Xiaomi MiMo-V2.5-Pro model cardchecked 2026-10-15 | Harness: 4-shot |
| MiMo-V2.5 BaseXiaomi, released 2026-04-27 | 67.7% | Xiaomi, Xiaomi MiMo-V2.5-Pro model cardchecked 2026-10-15 | Harness: 4-shot |
| MiMo-V2.5-Pro BaseXiaomi, released 2026-04-27 | 86.2% | Xiaomi, Xiaomi MiMo-V2.5-Pro model cardchecked 2026-10-15 | Harness: 4-shot |
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.