SWE-Bench Pro V2 HARD leaderboard
As of 16 Nov 2026, the top score on SWE-Bench Pro V2 HARD is 98.0% by Claude Opus 5 (Anthropic), ahead of Claude Fable 5.1 at 92.2%, across 18 model configurations with a published result (Scale AI SEAL, result dated 22 Sep 2026). The hardest subset of SWE-Bench Pro V2. Scale AI. SWE-Bench Pro V2 HARD is shown for reference only and does not count toward the BenchLeader Index.
Key facts (as of 16 Nov 2026)
- Claude Opus 5 holds the top SWE-Bench Pro V2 HARD score at 98.0% (Scale AI SEAL, 22 Sep 2026).
- Claude Fable 5.1 is second on SWE-Bench Pro V2 HARD at 92.2% (Scale AI SEAL, 22 Sep 2026).
- SWE-Bench Pro V2 HARD: The hardest subset of SWE-Bench Pro V2. Scale AI (Scale AI SEAL).
- 18 model configurations have a published SWE-Bench Pro V2 HARD result on BenchLeader (BenchLeader, 16 Nov 2026).
- SWE-Bench Pro V2 HARD does not feed the BenchLeader Index; it is shown for reference (BenchLeader methodology).
The hardest subset of SWE-Bench Pro V2. Scale AI.
- Published by
- Scale AI SEAL
- Category
- Coding
- Index weight
- Reference only
- Models
- 18
- Data as of
- 16 Nov 2026
- Newest result
- 6 Nov 2026
Scale AI SEAL Leaderboards.
- 1Claude Opus 5 (xhigh)98.0%
- 2Claude Fable 5.1 (high)92.2%
- 3Gemini 4 Argon (high)92.2%
- 4GPT-6 Astra (high)90.2%
- 5Muse Wonder (xhigh)90.2%
- 6Kimi K3 (max)88.2%
- 7Claude Sonnet 5 (xhigh)88.2%
- 8GPT-5.6 Terra (xhigh)86.3%
- 9Inkling 2 Small86.3%
- 10MiniMax M3.1 Flash (xhigh)84.3%
- 11GLM 5.3 (max)84.3%
- 12GPT-5.6 Sol (xhigh)82.4%
- 13Muse Spark Open (xhigh)82.3%
- 14Claude Sonnet 5.5 (xhigh)70.6%
- 15Gemini 3.8 Flash (high)58.8%
18 of 18
| # | ||||
|---|---|---|---|---|
| 1 | 98.0% | 60.1 | 2026-09-22 | |
| 2 | 92.2% | 63.7 | 2026-09-22 | |
| 3 | 92.2% | 65.3 | 2026-10-14 | |
| 4 | 90.2% | 66.3 | 2026-09-22 | |
| 5 | 90.2% | – | 2026-10-14 | |
| 6 | 88.2% | 60.2 | 2026-09-22 | |
| 7 | 88.2% | 55.4 | 2026-09-22 | |
| 8 | 86.3% | 60.1 | 2026-09-22 | |
| 9 | 86.3% | 57.9 | 2026-11-06 | |
| 10 | 84.3% | – | 2026-11-02 | |
| 11 | 84.3% | 58.8 | 2026-09-22 | |
| 12 | 82.4% | 63.9 | 2026-09-22 | |
| 13 | 82.3% | 53.5 | 2026-10-16 | |
| 14 | 70.6% | 62.5 | 2026-10-12 | |
| 15 | 58.8% | 59.4 | 2026-09-22 | |
| 16 | 56.9% | 50.0 | 2026-09-22 | |
| 17 | 52.9% | 55.3 | 2026-10-08 | |
| 18 | 25.5% | – | 2026-09-22 |
AI coding benchmark leaderboard · All LLM benchmarks
Related benchmarks
Top head-to-heads on SWE-Bench Pro V2 HARD
Cite as: BenchLeader, “SWE-Bench Pro V2 HARD leaderboard”, data as of 16 Nov 2026, https://www.benchleader.com/benchmarks/scale_swe_bench_pro_v2_hard.
SWE-Bench Pro V2 HARD: questions
- What does SWE-Bench Pro V2 HARD measure?
- The hardest subset of SWE-Bench Pro V2. Scale AI. Scores are reported in percent of tasks solved; higher is better.
- What is the top score on SWE-Bench Pro V2 HARD?
- As of 16 Nov 2026, the highest score on SWE-Bench Pro V2 HARD is 98.0% by Claude Opus 5 (result dated 22 Sep 2026), ahead of Claude Fable 5.1 at 92.2%. Scores come from Scale AI SEAL and are re-read daily.
- Which AI model leads SWE-Bench Pro V2 HARD?
- Claude Opus 5 leads SWE-Bench Pro V2 HARD with 98.0% as of 16 Nov 2026 (result dated 22 Sep 2026), ahead of Claude Fable 5.1 at 92.2%.
- How many models have SWE-Bench Pro V2 HARD results?
- 18 model configurations have a SWE-Bench Pro V2 HARD result on BenchLeader, all taken from Scale AI SEAL.
- Who runs SWE-Bench Pro V2 HARD and how often is it updated?
- SWE-Bench Pro V2 HARD is published by Scale AI SEAL. BenchLeader re-reads the published results once a day and records the date each result was published.
- Does SWE-Bench Pro V2 HARD count toward the BenchLeader Index?
- No. SWE-Bench Pro V2 HARD is shown for reference but left out of the composite index.