SWE-Bench Pro (private) leaderboard
As of 6 Nov 2026, the top score on SWE-Bench Pro (private) is 61.5% by Muse Spark 1.1 (Meta), ahead of GPT-5.4 at 59.1%, across 24 model configurations with a published result (Scale AI SEAL, result dated 9 Jul 2026). The same task format as SWE-Bench Pro on a held-out set of commercial repositories only Scale can run. SWE-Bench Pro (private) is shown for reference only and does not count toward the BenchLeader Index.
Key facts (as of 6 Nov 2026)
- Muse Spark 1.1 holds the top SWE-Bench Pro (private) score at 61.5% (Scale AI SEAL, 9 Jul 2026).
- GPT-5.4 is second on SWE-Bench Pro (private) at 59.1% (Scale AI SEAL, 8 Apr 2026).
- SWE-Bench Pro (private): The held-out commercial-repository set of SWE-Bench Pro. Scale AI (Scale AI SEAL).
- 24 model configurations have a published SWE-Bench Pro (private) result on BenchLeader (BenchLeader, 6 Nov 2026).
- SWE-Bench Pro (private) does not feed the BenchLeader Index; it is shown for reference (BenchLeader methodology).
The held-out commercial-repository set of SWE-Bench Pro. Scale AI.
- Published by
- Scale AI SEAL
- Category
- Coding
- Index weight
- Reference only
- Models
- 24
- Data as of
- 6 Nov 2026
Scale AI SEAL Leaderboards.
What the test looks like
The same task format as SWE-Bench Pro on a held-out set of commercial repositories only Scale can run.
How it is scored
Percent of issues resolved, published by Scale AI.
What to keep in mind
Contamination-proof but not reproducible outside Scale.
- 1Muse Spark 1.161.5%
- 2GPT-5.4 (xhigh)59.1%
- 3Muse Spark55.0%
- 4Claude Opus 4.6 (thinking)51.9%
- 5Gemini 3.1 Pro (thinking)46.1%
- 6Claude Opus 4.545.9%
- 7Claude Sonnet 4.543.6%
- 8Gemini 3 Pro43.3%
- 9Claude Sonnet 442.7%
- 10GPT-5 (high)41.8%
- 11GPT-5.2-Codex41.0%
- 12Claude Haiku 4.539.5%
- 13Qwen3 Coder 480B38.7%
- 14Minimax 2.136.8%
- 15Gemini 3 Flash34.6%
| # | ||||
|---|---|---|---|---|
| 1 |  Muse Spark 1.1 | 61.5% | 62.2 | 2026-07-09 |
| 2 |  GPT-5.4xhigh | 59.1% | 63.2 | 2026-04-08 |
| 3 | Muse Spark | 55.0% | 64.1 | 2026-04-08 |
| 4 |  Claude Opus 4.6thinking | 51.9% | 62.4 | 2026-04-08 |
| 5 |  Gemini 3.1 Prothinking | 46.1% | – | 2026-04-08 |
| 6 | Claude Opus 4.5 | 45.9% | 57.9 | 2025-12-11 |
| 7 | Claude Sonnet 4.5 | 43.6% | 53.7 | 2025-09-19 |
| 8 | Gemini 3 Pro | 43.3% | 60.2 | 2025-11-26 |
| 9 | Claude Sonnet 4 | 42.7% | 50.5 | 2025-09-19 |
| 10 | GPT-5high | 41.8% | 57.9 | 2025-11-26 |
| 11 | GPT-5.2-Codex | 41.0% | 57.3 | 2026-01-27 |
| 12 | Claude Haiku 4.5 | 39.5% | 47.6 | 2025-09-19 |
| 13 |  Qwen3 Coder 480Bopen ↗ | 38.7% | 48.8 | 2026-01-27 |
| 14 |  Minimax 2.1 | 36.8% | – | 2026-01-12 |
| 15 | Gemini 3 Flash | 34.6% | 56.9 | 2026-01-12 |
| 16 | GPT-5.2 | 29.9% | 57.9 | 2026-01-27 |
| 17 |  Kimi K2open ↗ | 27.7% | 51.0 | 2025-09-19 |
| 18 | Qwen3 235B A22Bopen ↗ | 21.4% | 50.0 | 2026-01-27 |
| 19 | gpt-oss-120bopen ↗ | 16.2% | 45.5 | 2026-01-27 |
| 20 | Gemma 3 27Bopen ↗ | 11.4% | 39.9 | 2026-01-27 |
| 21 | Llama 3.1 405Bopen ↗ | 11.2% | 41.9 | 2026-01-27 |
| 22 |  GLM-4.6open ↗ | 9.7% | 49.8 | 2026-01-27 |
| 23 | Llama 4 Maverickopen ↗ | 5.2% | 42.1 | 2026-01-27 |
| 24 |  Codestral 2405 | 1.5% | – | 2026-01-27 |
More coding benchmarks · All LLM benchmarks
Related benchmarks
Top head-to-heads on SWE-Bench Pro (private)
Cite as: BenchLeader, “SWE-Bench Pro (private) leaderboard”, data as of 6 Nov 2026, https://www.benchleader.com/benchmarks/scale_swe_bench_pro_private.
SWE-Bench Pro (private): questions
- What does SWE-Bench Pro (private) measure?
- The same task format as SWE-Bench Pro on a held-out set of commercial repositories only Scale can run. Scores are reported in percent of tasks solved; higher is better.
- What is the top score on SWE-Bench Pro (private)?
- As of 6 Nov 2026, the highest score on SWE-Bench Pro (private) is 61.5% by Muse Spark 1.1 (result dated 9 Jul 2026), ahead of GPT-5.4 at 59.1%. Scores come from Scale AI SEAL and are re-read daily.
- Which AI model leads SWE-Bench Pro (private)?
- Muse Spark 1.1 leads SWE-Bench Pro (private) with 61.5% as of 6 Nov 2026 (result dated 9 Jul 2026), ahead of GPT-5.4 at 59.1%.
- How many models have SWE-Bench Pro (private) results?
- 24 model configurations have a SWE-Bench Pro (private) result on BenchLeader, all taken from Scale AI SEAL.
- Who runs SWE-Bench Pro (private) and how often is it updated?
- SWE-Bench Pro (private) 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 (private) count toward the BenchLeader Index?
- No. SWE-Bench Pro (private) is shown for reference but left out of the composite index.