Coding Agent 能力天梯
基于 SWE-bench Verified 官方提交，衡量 AI 系统解决真实 GitHub 软件问题的能力。
方法版本 v1.0.0
来源复核 2026/7/19 18:29:58
源内最新评测 2026-02-26
| 排名 | 系统 / Agent | Resolved | 评测日期 | 证据 |
|---|---|---|---|---|
| #1 | live-SWE-agent + Claude 4.5 Opus medium (20251101) claude-opus-4-5-20251101 | 79.20% | 2025-12-15 | A 级 |
| #2 | Sonar Foundation Agent + Claude 4.5 Opus claude-opus-4-5 | 79.20% | 2025-12-05 | A 级 |
| #3 | TRAE + Doubao-Seed-Code Doubao-Seed-Code | 78.80% | 2025-09-28 | A 级 |
| #4 | live-SWE-agent + Gemini 3 Pro Preview (2025-11-18) gemini-3-pro-preview | 77.40% | 2025-11-20 | A 级 |
| #5 | Atlassian Rovo Dev (2025-09-02) claude-sonnet-4-20250514 | 76.80% | 2025-09-02 | A 级 |
| #6 | EPAM AI/Run Developer Agent v20250719 + Claude 4 Sonnet claude-sonnet-4-20250514 | 76.80% | 2025-08-04 | A 级 |
| #7 | mini-SWE-agent + Claude 4.5 Opus (high reasoning) claude-4-5-opus | 76.80% | 2026-02-17 | A 级 |
| #8 | ACoder claude-4-sonnet | 76.40% | 2025-08-19 | A 级 |
| #9 | mini-SWE-agent + Gemini 3 Flash (high reasoning) gemini-3-flash-preview | 75.80% | 2026-02-17 | A 级 |
| #10 | mini-SWE-agent + MiniMax M2.5 (high reasoning) minimax-m2.5 | 75.80% | 2026-02-17 | A 级 |
| #11 | mini-SWE-agent + Claude Opus 4.6 claude-opus-4-6 | 75.60% | 2026-02-17 | A 级 |
| #12 | Warp gpt-5 | 75.60% | 2025-09-01 | A 级 |
| #13 | TRAE + Claude Sonnet 4 + Opus 4 + Sonnet 3.7 + Gemini 2.5 Pro claude-4-sonnet-20250522 | 75.20% | 2025-06-12 | A 级 |
| #14 | Harness AI claude-sonnet-4-20250514 | 74.80% | 2025-07-31 | A 级 |
| #15 | Sonar Foundation Agent + Claude 4.5 Sonnet claude-sonnet-4-5 | 74.80% | 2025-11-03 | A 级 |
| #16 | JoyCode + Claude 4 Sonnet + GPT-4.1 claude-4-sonnet | 74.60% | 2025-09-15 | A 级 |
| #17 | Lingxi-v1.5_claude-4-sonnet-20250514 claude-4-sonnet-20250514 | 74.60% | 2025-07-20 | A 级 |
| #18 | mini-SWE-agent + Claude 4.5 Opus medium (20251101) claude-opus-4-5-20251101 | 74.40% | 2025-11-24 | A 级 |
| #19 | Prometheus-v1.2.1 + GPT-5 gpt-5-2025-08-07 | 74.40% | 2025-10-15 | A 级 |
| #20 | Refact.ai Agent + Claude 4 Sonnet + o4-mini claude-4-sonnet | 74.40% | 2025-06-03 | A 级 |
| #21 | mini-SWE-agent + Gemini 3 Pro Preview (2025-11-18) gemini-3-pro-preview | 74.20% | 2025-11-18 | A 级 |
| #22 | Salesforce AI Research SAGE (OpenHands) claude-sonnet-4.5 | 73.80% | 2025-11-03 | A 级 |
| #23 | Tools + Claude 4 Opus (2025-05-22) claude-4-opus-20250514 | 73.20% | 2025-05-22 | A 级 |
| #24 | Salesforce AI Research SAGE (bash-only) claude-sonnet-4.5 | 73.00% | 2025-10-21 | A 级 |
| #25 | mini-SWE-agent + GLM-5 (high reasoning) glm-5 | 72.80% | 2026-02-17 | A 级 |
| #26 | mini-SWE-agent + GPT-5-2 (high reasoning) gpt-5-2 | 72.80% | 2026-02-17 | A 级 |
| #27 | mini-SWE-agent + GPT-5-2 Codex gpt-5-2-codex | 72.80% | 2026-02-19 | A 级 |
| #28 | Tools + Claude 4 Sonnet (2025-05-22) claude-4-sonnet-20250514 | 72.40% | 2025-05-22 | A 级 |
| #29 | mini-SWE-agent + GPT-5.2 (2025-12-11) (high reasoning) gpt-5.2-2025-12-11 | 71.80% | 2025-12-11 | A 级 |
| #30 | OpenHands + GPT-5 openai/gpt-5-2025-08-07 | 71.80% | 2025-08-07 | A 级 |
| #31 | mini-SWE-agent + Claude 4.5 Sonnet (high reasoning) claude-sonnet-4-5-20250929 | 71.40% | 2026-02-17 | A 级 |
| #32 | Bloop claude-4-sonnet-20250514 | 71.20% | 2025-07-10 | A 级 |
| #33 | Lingxi v1.5 x Kimi K2 kimi-k2-0905-preview | 71.20% | 2025-10-14 | A 级 |
| #34 | Prometheus-v1.2 + GPT-5 gpt-5-2025-08-07 | 71.20% | 2025-09-29 | A 级 |
| #35 | Qodo Command claude-sonnet-4-20250514 | 71.20% | 2025-07-15 | A 级 |
| #36 | Warp claude-sonnet-4-20250514 | 71.00% | 2025-06-23 | A 级 |
| #37 | mini-SWE-agent + Kimi K2.5 (high reasoning) kimi-k2.5 | 70.80% | 2026-02-17 | A 级 |
| #38 | Moatless Tools + Claude 4 Sonnet claude-4-sonnet-20250514 | 70.80% | 2025-06-11 | A 级 |
| #39 | mini-SWE-agent + Claude 4.5 Sonnet (20250929) claude-sonnet-4-5-20250929 | 70.60% | 2025-09-29 | A 级 |
| #40 | TRAE | 70.60% | 2025-05-19 | A 级 |
| #41 | Augment Agent v1 | 70.40% | 2025-06-10 | A 级 |
| #42 | OpenHands + Claude 4 Sonnet claude-4-sonnet-20250514 | 70.40% | 2025-05-24 | A 级 |
| #43 | Refact.ai Agent claude-3-7-sonnet-20250219 | 70.40% | 2025-05-15 | A 级 |
| #44 | devlo claude-3-7-sonnet-20250219 | 70.20% | 2025-05-19 | A 级 |
| #45 | mini-SWE-agent + DeepSeek V3.2 (high reasoning) deepseek-v3.2 | 70.00% | 2026-02-17 | A 级 |
| #46 | Zencoder (2025-04-30) claude-3-7-sonnet-20250219 | 70.00% | 2025-04-30 | A 级 |
| #47 | mini-SWE-agent + Gemini 3 Pro gemini-3-pro-preview | 69.60% | 2026-02-26 | A 级 |
| #48 | OpenHands + Qwen3-Coder-480B-A35B-Instruct https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct | 69.60% | 2025-08-05 | A 级 |
| #49 | mini-SWE-agent + GPT-5.2 (2025-12-11) gpt-5.2-2025-12-11 | 69.00% | 2025-12-11 | A 级 |
| #50 | GLM-4.6 https://huggingface.co/zai-org/GLM-4.6 | 68.20% | 2025-09-30 | A 级 |
排名规则
按 SWE-bench 官方 Verified resolved 百分比降序排列；同分按名称排序。
仅收录官方实验仓库中有数值成绩的提交，展示前 50 名。
正确理解这个榜单
完整系统榜包含模型、Agent harness 和配置差异；页面保留官方标签，不把系统成绩误标为基础模型成绩。
商业合作、GitHub stars、社交热度和 AIUtil 编辑偏好不参与能力排名。