Benchmark
SWE-Bench Pro
Agentic benchmarks measure how well models carry out tasks that take several steps. Scores are percentages. Higher is better.
- Models and systems measured
- 93
- Scores recorded
- 188
- Sources
- 4
- IQ weight
- 3.48
Score distribution
- 57.3%
- Mean
- 57.7%
- Median
- 14%
- Std dev
- 46.3%
- 10th percentile
- 69.2%
- 90th percentile
- 5.2%
- Min
This benchmark has a weight of 3.48 for chat models, measured over 92 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model or system | Score |
|---|---|---|
| 51 | GPT-5.6 Terra· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 63.4% |
| 52 | GPT-5.6 TerraOpenAI GPT-5.6 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 63.4% |
| 53 | Qwen3.8 Omni FlashClaude Code · temp 1.0 · top_p 0.95 · 256K · refined tasks;Claude 对照为官方值· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 63.3% |
| 54 | Claude Sonnet 5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 63.2% |
| 55 | Claude Sonnet 5adaptive thinking · max effort · 5 trials· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 63.2% |
| 56 | Claude Sonnet 5Gemini 3.6 Flash Model Card · July 2026 · Public· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 63.2% |
| 57 | GPT-5.6 Luna· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 62.7% |
| 58 | GPT-5.6 LunaGemini 3.6 Flash Model Card · July 2026 · Public· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 62.7% |
| 59 | GPT-5.6 LunaOpenAI GPT-5.6 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 62.7% |
| 60 | Qwen 3.8 Flash Next· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 62.5% |
| 61 | Qwen 3.8 Flash NextQwen3.8-Flash-Next release/model card · Claude Code · temp 1.0 · top_p 0.95 · 256K · corrected/refined tasks· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 62.5% |
| 62 | Qwen3.8 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 62.5% |
| 63 | Qwen3.8 FlashClaude Code · temp 1.0 · top_p 0.95 · 256K · refined tasks;Claude 对照为官方值· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 62.5% |
| 64 | GLM 5.2· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 62.1% |
| 65 | GLM 5.2OpenHands · tailored prompt · 400K context · 32K max output· Reported by Z.ai· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 62.1% |
| 66 | GLM 5.2resolve rate· Reported by SpaceXAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 62.1% |
| 67 | Qwen 3.8 27B· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 61.7% |
| 68 | Qwen 3.8 27BClaude Code · temp 1.0 · top_p 0.95 · 256K · refined tasks;Claude 对照为官方值· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 61.7% |
| 69 | Qwen 3.8 27BQwen3.8-27B model card · Claude Code · temp 1.0 · top_p 0.95 · 256K · corrected/refined tasks· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 61.7% |
| 70 | Kimi K3Atria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · OpenHands v1.23.0 · 4.5-hour limit · 250 interaction steps · network disabled and Git history removed· Reported by Shanghai AI Laboratory· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 61.6% |
| 71 | Muse Spark 1.1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 61.5% |
| 72 | Muse Spark 1.1Meta Model API · xhigh · Scale AI · mini-swe-agent for Muse Spark 1.1· Reported by Meta, Meta Superintelligence Labs· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 61.5% |
| 73 | GPT-5.6 SolAtria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · OpenHands v1.23.0 · 4.5-hour limit · 250 interaction steps · network disabled and Git history removed· Reported by Shanghai AI Laboratory· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 61.4% |
| 74 | Qwen3.7 Max· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 60.6% |
| 75 | Qwen3.7 MaxQwen3.7-Max release · internal bash + file-edit agent scaffold · temp 1.0 · top_p 0.95 · 200K · corrected/refined tasks· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 60.6% |
| 76 | Qwen3.7 MaxQwen表:Claude Code · temp 1.0 · top_p 0.95 · 256K · refined tasks· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 60.6% |
| 77 | GLM 5.3Atria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · OpenHands v1.23.0 · 4.5-hour limit · 250 interaction steps · network disabled and Git history removed· Reported by Shanghai AI Laboratory· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 60.3% |
| 78 | LongCat 2.0· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim | 59.5% |
| 79 | GPT 5.5OpenAI GPT-5.6 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 59.4% |
| 80 | Laguna S 2.1· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 59.4% |
| 81 | MiniMax M3· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 59% |
| 82 | MiniMax M3Claude Code · official-aligned evaluation· Reported by MiniMax Research· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 59% |
| 83 | MiniMax M3official model-page overview figure· Reported by MiniMax· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 59% |
| 84 | Gemini 3.6 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 58.7% |
| 85 | Gemini 3.6 FlashGemini 3.6 Flash Model Card · July 2026 · Public· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 58.7% |
| 86 | GPT 5.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 58.6% |
| 87 | GPT 5.5Fable/Mythos 5 System Card capability summary · adaptive thinking · max effort · 5 trials unless noted· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 58.6% |
| 88 | GPT 5.5Gemini 3.5 Flash Model Card · May 2026 · Public · single attempt· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 58.6% |
| 89 | GPT 5.5Opus 4.8 System Card capability summary · adaptive thinking · max effort · 5 trials unless noted· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 58.6% |
| 90 | GPT 5.5Public · single pass · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 58.6% |
| 91 | GPT 5.5xhigh· Reported by SpaceXAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 58.6% |
| 92 | GPT 5.5xhigh · Scale AI · mini-swe-agent for Muse Spark 1.1· Reported by Meta, Meta Superintelligence Labs· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 58.6% |
| 93 | Kimi K2.6· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 58.6% |
| 94 | Kimi K2.6Kimi K2.6 official benchmark table · thinking · temp 1.0 · top_p 1.0 · 262,144 context · in-house SWE-agent-derived harness · avg@10· Reported by Moonshot AI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 58.6% |
| 95 | GLM 5.1· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 58.4% |
| 96 | GLM 5.1GLM-5.1 official release table· Reported by Z.ai· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 58.4% |
| 97 | DeepSeek V4 Pro 0813Atria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · OpenHands v1.23.0 · 4.5-hour limit · 250 interaction steps · network disabled and Git history removed· Reported by Shanghai AI Laboratory· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 58.3% |
| 98 | Hy3 (Deep Infra)· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 57.9% |
| 99 | Hy3 (Deep Infra)Hy3 official Benchmark Appendix · highest reasoning tier unless noted· Reported by Tencent Hunyuan· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 57.9% |
| 100 | GPT 5.4· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 57.7% |
Showing 51–100 of 187 results.
Not ranked
The source marks these results as descriptive, not capability scores, so we list them by name instead of ranking them.
| Model or system | Score |
|---|---|
| Muse Spark (version unspecified)Meta Model API · highest effort · version unspecified · Scale AI · mini-swe-agent for Muse Spark 1.1· Reported by Meta, Meta Superintelligence Labs· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 55% |
Data credits: Hugging Face official benchmark leaderboards; licence varies by benchmark · Data by LLM Stats · BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links. · OpenEvals/leaderboard-data, MIT (licence)