SuperGPQA: Scaling LLM Evaluation Across 285 Graduate Disciplines (SuperGPQA)
An expanded version of GPQA that evaluates graduate-level knowledge and reasoning capabilities across 285 disciplines, providing comprehensive coverage of academic domains.
Top models on SuperGPQA — October 29, 2026
As of October 29, 2026, Claude Opus 4.6 leads the SuperGPQA leaderboard with 95% , followed by Claude Sonnet 4.6 (95%) and Qwen 3.6 Max (preview) (73.9%).
Claude Opus 4.6
Anthropic
Claude Sonnet 4.6
Anthropic
Qwen 3.6 Max (preview)
Alibaba
24 modelsKnowledge2% of Knowledge reference weightCurrentUpdated October 29, 2026
Leaderboard (24 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Claude Opus 4.6Anthropic | 95% | Not reported | Closed |
| 2 | Claude Sonnet 4.6Anthropic | 95% | Not reported | Closed |
| 3 | Qwen 3.6 Max (preview)Alibaba | 73.9% | Not reported | Closed |
| 4 | Qwen3.7 MaxAlibaba | 73.6% | Not reported | Closed |
| 5 | Qwen3.6 PlusAlibaba | 71.6% | Not reported | Closed |
| 6 | Qwen3.7 PlusAlibaba | 71.4% | Not reported | Closed |
| 7 | Seed 2.1 ProByteDance | 70.8% | Not reported | Closed |
| 8 | Claude Opus 4.5Anthropic | 70.6% | Not reported | Closed |
| 9 | Qwen3.5 397BAlibaba | 70.4% | Not reported | Open |
| 10 | Kimi K2.5Moonshot AI | 69.2% | Not reported | Open |
| 11 | Seed 2.1 TurboByteDance | 67.4% | Not reported | Closed |
| 12 | Qwen3.5-122B-A10BAlibaba | 67.1% | Not reported | Open |
| 13 | GLM-5Z.AI | 66.8% | Not reported | Open |
| 14 | Qwen3.6-27BAlibaba | 66% | Not reported | Open |
| 15 | Qwen3.5-27BAlibaba | 65.6% | Not reported | Open |
| 16 | Qwen3 235B 2507 (Reasoning)Alibaba | 64.9% | Not reported | Open |
| 17 | Qwen3.6-35B-A3BAlibaba | 64.7% | Not reported | Open |
| 18 | Qwen3.5-35B-A3BAlibaba | 63.4% | Not reported | Open |
| 19 | Qwen3 235B 2507Alibaba | 62.6% | Not reported | Open |
| 20 | Kimi K2Moonshot AI | 57.2% | Not reported | Closed |
| 21 | DeepSeek V4 Pro BaseDeepSeek | 53.9% | Not reported | Open |
| 22 | DeepSeek V4 Flash BaseDeepSeek | 46.5% | Not reported | Open |
| 23 | MiniCPM5-2BOpenBMB | 40.8% | Not reported | Open |
| 24 | MiniCPM5-1BOpenBMB | 23.1% | Not reported | Open |
According to BenchLM.ai, Claude Opus 4.6 leads the SuperGPQA benchmark with a score of 95%, followed by Claude Sonnet 4.6 (95%) and Qwen 3.6 Max (preview) (73.9%). There is significant spread across the leaderboard, making this benchmark effective at differentiating model capabilities.
24 models have been evaluated on SuperGPQA. The benchmark falls in the Knowledge category. BenchAlign v5.8 gives SuperGPQA 2% of the Knowledge reference weight, so it moves the Knowledge leaderboard and the overall ranking. Reference weights are relative weights in the calibrated model, not fixed shares of a score.
About SuperGPQA
Year
2025
Tasks
285 disciplines
Format
Multiple choice questions
Difficulty
Graduate level
SuperGPQA significantly expands the scope of graduate-level evaluation by covering 285 disciplines compared to GPQA's focus on 3 subjects. It maintains the same rigorous standards while providing broader coverage of academic knowledge.
Freshness and provenance
Version
SuperGPQA 2025
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 SuperGPQA measure?
An expanded version of GPQA that evaluates graduate-level knowledge and reasoning capabilities across 285 disciplines, providing comprehensive coverage of academic domains.
Which model scores highest on SuperGPQA?
Claude Opus 4.6 by Anthropic currently leads with a score of 95% on SuperGPQA.
How many models are evaluated on SuperGPQA?
24 AI models have published results on SuperGPQA in the BenchLM catalog.
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 6,100+ subscribers.
One email each week. Unsubscribe anytime.