Which coding agents have public SWE-bench results?
Entries from the official SWE-bench leaderboard data file that name a product we track (or its open-source core). Scores are the percentage of tasks resolved, as published.
- Entries shown
- 15
- Products with an entry
- 3
- Leaderboard rows scanned
- 323
- Pulled
- Nov 4, 2026
| Board | Submission | Resolved | Date | Product |
|---|---|---|---|---|
| Multilingual | GPT 5.2 CodexSWE-agent | 66.3% | 2026-02-20 | Codex cloudModel in another harness, not the Codex cloud product |
| Verified | GPT 5.2 CodexSWE-agent | 72.8% | 2026-02-19 | Codex cloudModel in another harness, not the Codex cloud product |
| Verified | GPT 5.1 Codex (medium)SWE-agent | 66.0% | 2025-11-24 | Codex cloudModel in another harness, not the Codex cloud product |
| Lite | Factory Code Droid | 31.33% | 2024-06-17 | Factory Droids |
| Test | Factory Code Droid | 19.27% | 2024-06-17 | Factory Droids |
| Verified | Factory Code Droid | 37.0% | 2024-06-17 | Factory Droids |
| Verified | Google Jules + Gemini 2.0 Flash (v20241212-experimental)Google | 52.2% | 2024-12-12 | Jules |
| Lite | OpenHands + CodeAct v2.1 (claude-3-5-sonnet-20241022)OpenHands | 41.67% | 2024-10-25 | OpenHands Cloud |
| Lite | OpenHands + CodeAct v1.8OpenHands | 26.67% | 2024-07-25 | OpenHands Cloud |
| Multimodal | OpenHands-Versa (Claude-Sonnet 4)OpenHands | 34.43% | 2025-05-28 | OpenHands Cloud |
| Multimodal | OpenHands-Versa (Claude-3.7 Sonnet)OpenHands | 31.33% | 2025-05-09 | OpenHands Cloud |
| Test | OpenHands + CodeAct v2.1 (claude-3-5-sonnet-20241022)OpenHands | 29.38% | 2024-11-03 | OpenHands Cloud |
| Verified | OpenHands + GPT-5OpenHands | 71.8% | 2025-08-07 | OpenHands Cloud |
| Verified | OpenHands + Claude 4 SonnetOpenHands | 70.4% | 2025-05-24 | OpenHands Cloud |
| Verified | OpenHands + Qwen3-Coder-480B-A35B-InstructQwen | 69.6% | 2025-08-05 | OpenHands Cloud |
Source: SWE-bench leaderboards.json · checked 2026-11-04
No product entry
Devin, Claude Code on the web, Cursor Cloud Agents, GitHub Copilot cloud agent, Replit Agent, Lovable, Bolt.new, v0, Base44, Firebase Studio, Kiro, Amp have no entry under their own name on any SWE-bench board. Vendors often publish model scores in their own announcements; we only list entries on the official leaderboard so numbers are comparable.
Read with care
- Old entries (2024) were run with older models and are not a guide to the product today.
- A model entry (such as GPT 5.2 Codex in mini-SWE-agent) measures the model in a neutral harness, not the vendor's cloud product.
- Terminal-Bench's leaderboard renders in JavaScript and was not readable by our fetcher, so it is not included yet.
Weekly, free
Which coding agent changed this week
One short email when coding agents and app builders change prices, limits, MCP support or benchmark results. Every item links its source.