SWE-bench Verified Leaderboard
coding52 models · Updated October 2026
On the SWE-bench Verified benchmark, Claude Opus 4.5 ranks #1 with a score of 79.2, while Qwen3 Coder 30B A3B Instruct offers the best score-per-dollar at $0.28/1M output tokens. The full ranking, with cost per million tokens, is below.
💰 Best value
Qwen3 Coder 30B A3B Instruct — score 60.4 at $0.28/1M output tokens
| # | Model | Score | Output / 1M | Context |
|---|---|---|---|---|
| 1 | Claude Opus 4.5Anthropic | 79.2 | $25.00 | 200K |
| 2 | Doubao-Seed-CodeZenMux | 78.8 | $1.12 | 256K |
| 3 | Gemini 3 Pro PreviewGoogle | 77.4 | $12.00 | 1M |
| 4 | Claude Sonnet 4Anthropic | 76.8 | $15.00 | 200K |
| 5 | Claude 4.5 OpusAI/ML API | 76.8 | $34.39 | 200K |
| 6 | Claude 4 SonnetNanoGPT | 76.4 | $14.99 | 200K |
| 7 | Gemini 3 Flash PreviewGoogle | 75.8 | $3.00 | 1M |
| 8 | MiniMax M2.5MiniMax | 75.8 | $0.95 | 205K |
| 9 | GPT-5OpenAI | 75.6 | $10.00 | 400K |
| 10 | Claude Opus 4.6Anthropic | 75.6 | $25.00 | 1M |
| 11 | Claude Sonnet 4.5Anthropic | 74.8 | $15.00 | 1M |
| 12 | Claude 4 OpusNanoGPT | 73.2 | $75.00 | 200K |
| 13 | GLM 5Z.ai | 72.8 | $3.20 | 205K |
| 14 | GPT-5.2OpenAI | 72.8 | $14.00 | 400K |
| 15 | GPT-5.2-CodexOpenAI | 72.8 | $14.00 | 400K |
| 16 | kimi-k2-0905-preview302.AI | 71.2 | $2.53 | 262K |
| 17 | Kimi K2.5MoonshotAI | 70.8 | $3.00 | 262K |
| 18 | Claude 3.7 SonnetHelicone | 70.4 | $15.00 | 200K |
| 19 | DeepSeek V3.2DeepSeek | 70.0 | $0.80 | 164K |
| 20 | Qwen3-Coder 480B-A35B InstructAlibaba | 69.6 | $7.50 | 262K |
| 21 | GLM 4.6Z.ai | 68.2 | $2.00 | 205K |
| 22 | Claude Haiku 4.5Anthropic | 66.6 | $5.00 | 200K |
| 23 | GPT-5.1OpenAI | 66.0 | $10.00 | 400K |
| 24 | GPT-5.1-CodexOpenAI | 66.0 | $10.00 | 400K |
| 25 | o1-previewOpenAI | 64.6 | — | — |
| 26 | o4 MiniOpenAI | 64.6 | $4.40 | 200K |
| 27 | GLM 4.5Z.ai | 64.2 | $2.20 | 131K |
| 28 | Claude 3.5 SonnetDigitalOcean | 63.4 | $15.00 | 200K |
| 29 | Kimi K2 0711MoonshotAI | 63.4 | $2.30 | 131K |
| 30 | MiniMax M2MiniMax | 61.0 | $1.02 | 205K |
| 31 | Qwen3 Coder 30B A3B InstructQwen | 60.4 | $0.28 | 262K |
| 32 | GPT-5 MiniOpenAI | 59.8 | $2.00 | 400K |
| 33 | o3OpenAI | 58.4 | $8.00 | 200K |
| 34 | Devstral Small 2505Mistral | 56.4 | $0.30 | 128K |
| 35 | Devstral MediumMistral | 53.8 | $2.00 | 128K |
| 36 | Gemini 2.5 Pro (Jun 2025)Google DeepMind | 53.6 | — | — |
| 37 | Kimi K2 InstructCortecs | 53.4 | $2.65 | 131K |
| 38 | Gemini 2.0 FlashGoogle | 52.2 | $0.40 | 1M |
| 39 | Qwen2.5 Coder 32B Instruct | 47.0 | $1.00 | 33K |
| 40 | o3 MiniOpenAI | 42.4 | $4.40 | 200K |
| 41 | DeepSeek V3DeepSeek | 42.0 | $0.28 | 164K |
| 42 | Claude 3 HaikuAnthropic | 40.6 | $1.25 | 200K |
| 43 | Qwen 2.5 MaxNanoGPT | 40.2 | $6.39 | 32K |
| 44 | GPT-4.1OpenAI | 39.6 | $8.00 | 1M |
| 45 | GPT-4oOpenAI | 38.8 | $10.00 | 128K |
| 46 | GPT-5 NanoOpenAI | 34.8 | $0.40 | 400K |
| 47 | Gemini 2.5 FlashGoogle | 28.7 | $2.50 | 1M |
| 48 | gpt-oss-120b (batch)OpenAI | 26.0 | $0.14 | 131K |
| 49 | GPT-4.1 MiniOpenAI | 23.9 | $1.60 | 1M |
| 50 | Claude 3 OpusLLM Gateway | 15.8 | $75.00 | 200K |
| 51 | Claude 2Anthropic | 4.4 | — | — |
| 52 | GPT-3.5 (davinci-002)OpenAI | 0.4 | — | — |
What does SWE-bench Verified test?
Evaluation framework with 2,294 real GitHub software-engineering problems.
Frequently asked questions
What is the SWE-bench Verified benchmark?
Evaluation framework with 2,294 real GitHub software-engineering problems.
Which model is best on SWE-bench Verified?
Claude Opus 4.5 (Anthropic) currently leads the SWE-bench Verified leaderboard with a score of 79.2.
What is the best value model on SWE-bench Verified?
Qwen3 Coder 30B A3B Instruct offers the best score-per-dollar on SWE-bench Verified — a score of 60.4 at $0.28 per 1M output tokens.
Pricing is indicative — confirm with the provider before production use. Updated October 2026.