Exgentic Open Agent - SWE-bench: leaderboard
Exgentic Open Agent SWE-bench evaluates agent frameworks and models on issue-resolution tasks from software repositories.
Metric: Success Rate (%). Source: www.exgentic.ai. 25 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | OpenAI Solo + claude-opus-4.5 | 80.72 | |
| 2 | Smolagent + gemini-3-pro | 75.76 | |
| 3 | Claude Code + claude-opus-4.5 | 74.23 | |
| 4 | OpenAI Solo + deepseek-v3.2 | 73.68 | |
| 5 | OpenAI Solo + gemini-3-pro | 72.34 | |
| 6 | React + gemini-3-pro | 71 | |
| 7 | React + Shortlisting + gemini-3-pro | 71 | |
| 8 | React + deepseek-v3.2 | 68.75 | |
| 9 | React + Shortlisting + deepseek-v3.2 | 68.75 | |
| 10 | Claude Code + gemini-3-pro | 67 |
Interactive version: theaggregate.ai/benchmark?slug=exgentic-open-agent-swe-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.