Market analysis
Updated
Benchmarks
Fixing real bugs in real GitHub projects, checked by the projects' own tests.
SWE-Bench Verified leaderboard
120 models · Score out of 100% · Higher is better · Status shows whether a score was verified, independently measured or self-reported
20 of 120 models in the chart · ·
All models
| Compare | # | Model | Creator | Status | Score |
|---|---|---|---|---|---|
| 1 | Claude Fable 5 |   Anthropic | Self-reported | 95.0% | |
| 2 | Claude Mythos Preview | Anthropic | Self-reported | 93.9% | |
| 3 | Ember-1 |  Fireworks AI | Self-reported | 92.2% | |
| 4 | Claude Opus 4.8 | Anthropic | Self-reported | 88.6% | |
| 5 | Claude Opus 4.7 | Anthropic | Self-reported | 87.6% | |
| 6 | Claude Sonnet 5 | Anthropic | Self-reported | 85.2% | |
| 7 | Claude Opus 4.5 | Anthropic | Self-reported | 80.9% | |
| 8 | Claude Opus 4.6 | Anthropic | Self-reported | 80.8% | |
| 9 | DeepSeek-V4-Pro-Max |  DeepSeek | Self-reported | 80.6% | |
| 10 | Gemini 3.1 Pro |  Google | Self-reported | 80.6% | |
| 11 | MiniMax M3 |  MiniMax | Self-reported | 80.5% | |
| 12 | Qwen3.7 Max |  Qwen | Self-reported | 80.4% | |
| 13 | Inkling-Small | Thinking Machines Lab | Self-reported | 80.2% | |
| 14 | Kimi K2.6 |   Moonshot AI | Self-reported | 80.2% | |
| 15 | MiniMax M2.5 | MiniMax | Self-reported | 80.2% | |
| 16 | GPT-5.2 |   OpenAI | Self-reported | 80.0% | |
| 17 | Claude Sonnet 4.6 | Anthropic | Self-reported | 79.6% | |
| 18 | DeepSeek-V4-Flash-Max | DeepSeek | Self-reported | 79.0% | |
| 19 | MiMo-V2.5-Pro |  Xiaomi | Self-reported | 78.9% | |
| 20 | Qwen3.6 Plus | Qwen | Self-reported | 78.8% | |
| 21 | DeepSeek-V4-Flash-0423 | DeepSeek | Self-reported | 78.6% | |
| 22 | Gemini 3 Flash | Google | Self-reported | 78.0% | |
| 23 | Hy3 |  Tencent | Self-reported | 78.0% | |
| 24 | MiMo-V2-Pro | Xiaomi | Self-reported | 78.0% | |
| 25 | GLM-5 |   Z.ai | Self-reported | 77.8% | |
| 26 | Qwen3.7-Plus | Qwen | Self-reported | 77.7% | |
| 27 | Inkling | Thinking Machines Lab | Self-reported | 77.6% | |
| 28 | Mistral Medium 3.5 |  Mistral | Self-reported | 77.6% | |
| 29 | Muse Spark |  Meta | Self-reported | 77.4% | |
| 30 | Qwen3.6-27B | Qwen | Self-reported | 77.2% | |
| 31 | Kimi K2.5 | Moonshot AI | Self-reported | 76.8% | |
| 32 | Seed 2.0 Pro |  ByteDance | Self-reported | 76.5% | |
| 33 | Qwen3.5-397B-A17B | Qwen | Self-reported | 76.4% | |
| 34 | GPT-5.1 | OpenAI | Self-reported | 76.3% | |
| 35 | GPT-5.1 Instant | OpenAI | Self-reported | 76.3% | |
| 36 | GPT-5.1 Thinking | OpenAI | Self-reported | 76.3% | |
| 37 | Gemini 3 Pro | Google | Self-reported | 76.2% | |
| 38 | Muse Glimmer-30B | Meta | Self-reported | 76.0% | |
| 39 | Qwen3 Max Thinking | Qwen | Self-reported | 75.3% | |
| 40 | GPT-5 | OpenAI | Self-reported | 74.9% | |
| 41 | MiMo-V2-Omni | Xiaomi | Self-reported | 74.8% | |
| 42 | Claude Opus 4.1 | Anthropic | Self-reported | 74.5% | |
| 43 | GPT-5 Codex | OpenAI | Self-reported | 74.5% | |
| 44 | Step-3.5-Flash | StepFun | Self-reported | 74.4% | |
| 45 | GLM-4.7 | Z.ai | Self-reported | 73.8% | |
| 46 | GPT-5.1 Codex | OpenAI | Self-reported | 73.7% | |
| 47 | MAI-Thinking-1 |  Microsoft | Self-reported | 73.5% | |
| 48 | Seed 2.0 Lite | ByteDance | Self-reported | 73.5% | |
| 49 | MiMo-V2-Flash | Xiaomi | Self-reported | 73.4% | |
| 50 | Qwen3.6-35B-A3B | Qwen | Self-reported | 73.4% | |
| 51 | Claude Haiku 4.5 | Anthropic | Self-reported | 73.3% | |
| 52 | DeepSeek-V3.2 | DeepSeek | Self-reported | 73.1% | |
| 53 | DeepSeek-V3.2 (Thinking) | DeepSeek | Self-reported | 73.1% | |
| 54 | DeepSeek-V3.2-Speciale | DeepSeek | Self-reported | 73.1% | |
| 55 | Seed 1.8 | ByteDance | Self-reported | 72.9% | |
| 56 | Claude Sonnet 4 | Anthropic | Self-reported | 72.7% | |
| 57 | MAI-Code-1.1-Flash | Microsoft | Self-reported | 72.6% | |
| 58 | Claude Opus 4 | Anthropic | Self-reported | 72.5% | |
| 59 | Qwen3.5-27B | Qwen | Self-reported | 72.4% | |
| 60 | U2 |  Unisound | Verified | 72.2% | |
| 61 | Qwen3.5-122B-A10B | Qwen | Self-reported | 72.0% | |
| 62 | MAI-Code-1-Flash | Microsoft | Self-reported | 71.6% | |
| 63 | Kimi K2-Thinking-0905 | Moonshot AI | Self-reported | 71.3% | |
| 64 | Nemotron 3.5 Super VL (120B A12B) |  NVIDIA | Self-reported | 71.1% | |
| 65 | Laguna XS 2.1 |  Poolside | Self-reported | 70.9% | |
| 66 | Grok Code Fast 1 |   xAI | Self-reported | 70.8% | |
| 67 | Nemotron 3 Ultra (550B A55B) | NVIDIA | Self-reported | 70.7% | |
| 68 | Solar Pro 4 | Upstage | Self-reported | 70.6% | |
| 69 | Claude 3.7 Sonnet | Anthropic | Self-reported | 70.3% | |
| 70 | LongCat-Flash-Thinking-2601 |  Meituan | Self-reported | 70.0% | |
| 71 | Nova 2 Pro |  Amazon | Self-reported | 70.0% | |
| 72 | Qwen3 Max | Qwen | Self-reported | 69.6% | |
| 73 | Qwen3-Coder 480B A35B Instruct | Qwen | Self-reported | 69.6% | |
| 74 | MiniMax M2 | MiniMax | Self-reported | 69.4% | |
| 75 | Qwen3.5-35B-A3B | Qwen | Self-reported | 69.2% | |
| 76 | o3 | OpenAI | Self-reported | 69.1% | |
| 77 | o4-mini | OpenAI | Self-reported | 68.1% | |
| 78 | GLM-4.6 | Z.ai | Self-reported | 68.0% | |
| 79 | DeepSeek-V3.2-Exp | DeepSeek | Self-reported | 67.8% | |
| 80 | North Mini Code 1.0 |  Cohere | Self-reported | 67.6% | |
| 81 | Gemini 2.5 Pro Preview 06-05 | Google | Self-reported | 67.2% | |
| 82 | MiniMax M2.1 | MiniMax | Self-reported | 67.0% | |
| 83 | DeepSeek-V3.1 | DeepSeek | Self-reported | 66.0% | |
| 84 | Kimi K2-Instruct-0905 | Moonshot AI | Self-reported | 65.8% | |
| 85 | Nova 2 Lite | Amazon | Self-reported | 64.5% | |
| 86 | GLM-4.5 | Z.ai | Self-reported | 64.2% | |
| 87 | Gemini 2.5 Pro | Google | Self-reported | 63.2% | |
| 88 | Devstral Medium | Mistral | Self-reported | 61.6% | |
| 89 | Gemini 2.5 Flash | Google | Self-reported | 60.4% | |
| 90 | LongCat-Flash-Chat | Meituan | Self-reported | 60.4% | |
| 91 | Nemotron Labs Dory 1.0 (4B) | NVIDIA | Self-reported | 59.5% | |
| 92 | LongCat-Flash-Thinking | Meituan | Self-reported | 59.4% | |
| 93 | GLM-4.7-Flash | Z.ai | Self-reported | 59.2% | |
| 94 | GLM-4.5-Air | Z.ai | Self-reported | 57.6% | |
| 95 | IBM Granite 4.2 30B |  IBM | Self-reported | 57.0% | |
| 96 | MiniMax M1 80K | MiniMax | Self-reported | 56.0% | |
| 97 | MiniMax M1 40K | MiniMax | Self-reported | 55.6% | |
| 98 | GPT-4.1 | OpenAI | Self-reported | 54.6% | |
| 99 | LongCat-Flash-Lite | Meituan | Self-reported | 54.4% | |
| 100 | Nemotron 3 Super (120B A12B) | NVIDIA | Self-reported | 53.7% | |
| 101 | Devstral Small 1.1 | Mistral | Self-reported | 53.6% | |
| 102 | Nemotron 3.5 Lightning (30B A3B) | NVIDIA | Self-reported | 51.6% | |
| 103 | o3-mini | OpenAI | Self-reported | 49.3% | |
| 104 | Claude 3.5 Sonnet | Anthropic | Self-reported | 49.0% | |
| 105 | IBM Granite 4.2 8B | IBM | Self-reported | 47.7% | |
| 106 | Sarvam-105B | Sarvam AI | Self-reported | 45.0% | |
| 107 | DeepSeek-R1-0528 | DeepSeek | Self-reported | 44.6% | |
| 108 | DeepSeek-V3 | DeepSeek | Self-reported | 42.0% | |
| 109 | o1-preview | OpenAI | Self-reported | 41.3% | |
| 110 | o1 | OpenAI | Self-reported | 41.0% | |
| 111 | Claude 3.5 Haiku | Anthropic | Self-reported | 40.6% | |
| 112 | Nemotron 3 Nano (30B A3B) | NVIDIA | Self-reported | 38.8% | |
| 113 | GPT-4.5 | OpenAI | Self-reported | 38.0% | |
| 114 | Sarvam-30B | Sarvam AI | Self-reported | 34.0% | |
| 115 | GPT-4o | OpenAI | Self-reported | 33.2% | |
| 116 | Gemini 2.5 Flash-Lite | Google | Self-reported | 31.6% | |
| 117 | GPT-4.1 mini | OpenAI | Self-reported | 23.6% | |
| 118 | Gemini Diffusion | Google | Self-reported | 22.9% | |
| 119 | DeepSeek-V2.5 | DeepSeek | Self-reported | 16.8% | |
| 120 | GPT-4o mini | OpenAI | Self-reported | 8.7% |
Score vs. price
How much you pay for each point of performance. Up and to the left is better value; highlighted models have the best score at or below their price.
GPQA score vs. input price
Top 60 models by score with a listed price · Lowest listed price per million input tokens