ArXivMath August 2026 without tools (ArXivMath Aug. 2026 (no tools))
We show this table for reference; we do not rank on it.
Final-answer research mathematics problems drawn from recent arXiv papers, including problems based on resolved or refuted conjectures.
Benchmark score on ArXivMath Aug. 2026 (no tools) — November 13, 2026
We compile the ArXivMath Aug. 2026 (no tools) rows from provider self-reports. Claude Opus 5.5 leads the table at 91.2%, followed by Claude Sonnet 5.5 (86.8%). We do not use these results to rank models overall.
Claude Opus 5.5
Anthropic
Reported by AnthropicClaude Sonnet 5.5
Anthropic
Reported by Anthropic2 modelsMathematicsCurrentDisplay onlyUpdated November 13, 2026
Benchmark score table (2 models)
All 2 results on this page were reported by the labs that built the models.
| Rank | Model / configuration | Score | Who ran it | Parameters (B) | Open / closed |
|---|---|---|---|---|---|
| 1 | Claude Opus 5.5AnthropicReported by Anthropic | 91.2% | Reported by Anthropic | Not reported | Closed |
| 2 | Claude Sonnet 5.5AnthropicReported by Anthropic | 86.8% | Reported by Anthropic | Not reported | Closed |
- Lab-reported
- Benchmark-run
- Independent
ArXivMath Aug. 2026 (tools)
Section 8.9 reports four-attempt average accuracy on the distinct 57-problem August 2026 release at max effort with tools. It remains separate from June and from MathArena's vendor-harness leaderboard.
| Source rank | Source model or configuration | Score | Who ran it | Source setup | Parameters (B) | Open / closed |
|---|---|---|---|---|---|---|
| 1 | Claude Opus 5.5Anthropic | 96.9% | Reported by Anthropic | MathArena and AnthropicOriginal result | Not reported | Closed |
| 2 | Claude Sonnet 5.5Anthropic | 95.2% | Reported by Anthropic | MathArena and AnthropicOriginal result | Not reported | Closed |
ArXivMath Jun. 2026 (no tools)
Section 8.8 reports four-run average accuracy on the 49-problem June 2026 release at max effort without tools.
| Source rank | Source model or configuration | Score | Who ran it | Source setup | Parameters (B) | Open / closed |
|---|---|---|---|---|---|---|
| 1 | Claude Opus 5Anthropic | 90.8% | Reported by Anthropic | MathArena and AnthropicOriginal result | Not reported | Closed |
ArXivMath Jun. 2026 (tools)
Section 8.8 reports four-run average accuracy on the 49-problem June 2026 release at max effort with tools.
| Source rank | Source model or configuration | Score | Who ran it | Source setup | Parameters (B) | Open / closed |
|---|---|---|---|---|---|---|
| 1 | Claude Opus 5Anthropic | 91.3% | Reported by Anthropic | MathArena and AnthropicOriginal result | Not reported | Closed |
About ArXivMath Aug. 2026 (no tools)
Year
2026
Tasks
57 recent research-mathematics problems
Format
Final-answer accuracy
Difficulty
Research mathematics
Section 8.9 reports four-attempt average accuracy on the distinct 57-problem August 2026 release at max effort without tools. It is not comparable to the June release and remains display-only.
Freshness and provenance
Version
ArXivMath Aug. 2026 (no tools) 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does ArXivMath Aug. 2026 (no tools) measure?
Final-answer research mathematics problems drawn from recent arXiv papers, including problems based on resolved or refuted conjectures.
Which model scores highest on ArXivMath Aug. 2026 (no tools)?
Claude Opus 5.5 by Anthropic currently leads with a score of 91.2% on ArXivMath Aug. 2026 (no tools).
How many models are evaluated on ArXivMath Aug. 2026 (no tools)?
2 AI models have published results on ArXivMath Aug. 2026 (no tools) in the BenchLM catalog.
Compare top models on ArXivMath Aug. 2026 (no tools)
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 7,000+ subscribers.
One email each week. Unsubscribe anytime.