Best LLMs for Math — October 2026 Leaderboard
Data refreshed:
Mathematical reasoning and problem solving
As of October 2026, the top math model on the BenchLM leaderboard is Claude Opus 4.8 with a weighted math score of 84.4.
Decision lens: use provisional-ranked mode for broader public evidence and verified-ranked mode for source-only comparisons. A model can move between views as evidence coverage changes.
- Data refreshed
- October 20, 2026
- Provisional-ranked
- 12 of 959 models
- Verified-ranked
- 11 of 959 models
- Weighted evidence
- 5 of 14 benchmarks
14 tracked benchmarks
FrontierMath v2 (Tiers 1-3), AIME26, HMMT Feb 2026, USAMO 2026, FrontierMath v2 (Tier 4), AIME 2023, AIME 2024, AIME 2025, AIME25 (Arcee), HMMT Feb 2023, HMMT Feb 2024, HMMT Feb 2025, BRUMO 2025, MATH-500
Evidence set: FrontierMath v2 (Tiers 1-3), AIME26, HMMT Feb 2026, USAMO 2026, FrontierMath v2 (Tier 4), AIME 2023, AIME 2024, AIME 2025, AIME25 (Arcee), HMMT Feb 2023, HMMT Feb 2024, HMMT Feb 2025, BRUMO 2025, MATH-500
Best Math picks
BenchLM summaries for math plus the practical tradeoffs users check next: open weights, price, speed, latency, and context.
Math Leaderboard
Primary score: weighted math score. Higher values rank first. Use the Show metric control to change the value shown in each row.
Switch between provisional-ranked and verified-ranked modes to compare the broader public dataset with sourced-only rankings.
Filters
| Parameters (B) | ||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
1 Claude Opus 4.8 Anthropic | Not reported | Closed | 84.4% | 77 | 80.000% | — | — | 96.7% | 56.098% | — | — | — | — | — | — | — | — | — |
2 Qwen3.7 Max Alibaba | Not reported | Closed | 79% | 76 | 64.561% | — | 97.1% | — | 34.146% | — | — | — | — | — | — | — | — | — |
| Not reported | Open | 78.4% | 65 | 59.206% | 99.2% | 92.5% | — | 29.268% | — | — | — | — | — | — | — | — | — | |
| Not reported | Open | 76.8% | 67 | 57.193% | 96.4% | 92.7% | — | 25.641% | — | — | — | — | — | — | — | — | — | |
5 | Not reported | Open | 69.8% | 64 | 46.316% | 95.5% | 90.2% | — | 17.073% | — | — | — | — | — | — | — | — | — |
6 Qwen3.6 Plus Alibaba | Not reported | Closed | 69.5% | 68 | 38.246% | 95.3% | 87.8% | — | 8.333%P | — | — | — | — | — | — | — | — | — |
| Not reported | Open | 66.7% | 64 | 36.842% | 95.3% | 82.6% | — | 12.500%P | — | — | 93.3%P | — | — | — | — | 87%P | 97.4%P | |
8 | Not reported | Open | 65.9% | 62 | 35.088% | 94.1% | 84.3% | — | — | — | — | — | — | — | — | — | — | — |
9 Claude Opus 4.5 Anthropic | Not reported | Closed | 61.3% | 60 | 34.386% | 95.1% | 85.3% | — | 4.878% | 99%P | 99%P | 98%P | — | 95%P | 97%P | 96%P | 96%P | 89%P |
10 | Not reported | Open | 58.4% | 55 | 20.351% | 92.7% | 83.6% | — | — | — | — | — | — | — | — | — | — | — |
| Not reported | Open | 58.3% | 67 | 33.333% | 97.1% | — | — | 4.878% | — | — | — | — | — | — | — | — | — | |
12 Kimi K2 Moonshot AI | Not reported | Closed | 0% | 52 | 21.404%P | — | — | — | 0.000%P | — | 69.6% | 49.5% | — | — | — | 38.8%P | — | 97.4% |
Top AI models for Math — October 2026
As of October 2026, Claude Opus 4.8 leads the provisional math leaderboard with a score of 84.4%, followed by Qwen3.7 Max (79.0%) and GLM-5.2 (78.4%). BenchLM is currently showing 12 provisional-ranked models and 11 verified-ranked models in this category.
Scored on 3 of the 5 weighted math benchmarks.
Scored on 3 of the 5 weighted math benchmarks.
Scored on 4 of the 5 weighted math benchmarks.
What changed
Claude Opus 4.8 ranks #1 at 84.4 on the weighted math score.
Qwen3.7 Max ranks #2 at 79.0 on the weighted math score.
GLM-5.2 ranks #3 at 78.4 on the weighted math score.
Top models by benchmark
A 2026 American Invitational Mathematics Examination snapshot used in frontier-model comparison tables for mathematical reasoning.(25% of category score)
Score in Context
What these scores mean
The Math leaderboard ranks models by a weighted category score. The overall ranking combines external indices with benchmark evidence rather than fixed category weights. A model needs sourced results on 3 of the 5 weighted benchmarks to enter this table. We leave a model with fewer off the table; its model page still shows the results it has.
Known limitations
AIME 2023, AIME 2024, AIME 2025, AIME25 (Arcee), HMMT Feb 2023, HMMT Feb 2024, HMMT Feb 2025, BRUMO 2025, and MATH-500 stay listed for reference and carry no weight. Contest sets saturate quickly at the frontier and older problems can reach training data, so the weighted sets rotate each year. See the AIME & HMMT explainer.
How we weight
The math score weights FrontierMath v2 (Tiers 1-3) at 30%, AIME26 at 25%, HMMT Feb 2026 at 25%, USAMO 2026 at 10%, and FrontierMath v2 (Tier 4) at 10%.
See the math leaderboard or read the AIME & HMMT explainer.
Leaderboards exclude benchmark rows that BenchLM generated from other scores or cloned from reference models. When a weighted benchmark is missing after that filter, the category falls back to the remaining trustworthy public rows instead of filling the gap with synthetic values.
The full scoring rules, freshness handling, and runtime/pricing caveats live on the BenchLM methodology page.
Scroll horizontally to read the full evidence ledger.
| Benchmark | Weight | Status | Description |
|---|---|---|---|
| FrontierMath v2 (Tiers 1-3) | 30% | Weighted | Epoch AI's corrected v2 core FrontierMath suite of private advanced mathematics problems. Models can reason iteratively and use Python; scores are pass rates on the private set. |
| AIME26 | 25% | Weighted | A 2026 American Invitational Mathematics Examination snapshot used in frontier-model comparison tables for mathematical reasoning. |
| HMMT Feb 2026 | 25% | Weighted | A February 2026 HMMT slice used in newer frontier-model math comparisons. |
| USAMO 2026 | 10% | Weighted | The premier US mathematical olympiad competition, featuring proof-based problems that require deep mathematical insight and rigorous argumentation at the highest competition level. |
| FrontierMath v2 (Tier 4) | 10% | Weighted | Epoch AI's corrected v2 Tier 4 expansion, a separate set of exceptionally difficult research-level mathematics problems evaluated with Python-enabled iterative reasoning. |
| AIME 2023 | — | Display only | High school mathematics competition |
| AIME 2024 | — | Display only | High school mathematics competition |
| AIME 2025 | — | Display only | High school mathematics competition |
| AIME25 (Arcee) | — | Display only | Display-only AIME25 reference from Arcee AI's Trinity-Large-Thinking launch chart. |
| HMMT Feb 2023 | — | Display only | Collegiate mathematics competition |
| HMMT Feb 2024 | — | Display only | Collegiate mathematics competition |
| HMMT Feb 2025 | — | Display only | Collegiate mathematics competition |
| BRUMO 2025 | — | Display only | University-level mathematics olympiad |
| MATH-500 | — | Display only | Curated 500-problem subset of the MATH dataset covering algebra, geometry, number theory, and more |
About Math benchmarks
Epoch AI's corrected v2 core FrontierMath suite of private advanced mathematics problems. Models can reason iteratively and use Python; scores are pass rates on the private set.
Questions
What is the best LLM for math?
The best LLMs for math are ranked by competition-level benchmarks like AIME and HMMT, with top models achieving strong scores on problems from real math competitions.
How are math benchmarks scored for LLMs?
Math benchmarks score models on correctness of final answers to problems ranging from algebra to advanced competition mathematics, testing both calculation and mathematical reasoning.
What benchmarks test math ability in AI models?
Major math benchmarks include AIME (competition-level problems), HMMT (Harvard-MIT tournament problems), and BRUMO, each testing progressively harder mathematical reasoning.
Math benchmark updates
Math model rankings change weekly. Stay current.
Read a sample issueJoin 5,600+ readers.
One email each week. Unsubscribe anytime.