# SWE-Bench Pro

Tests whether AI agents can solve long, realistic software tasks in real open-source projects, harder than older bug-fixing tests.

Category: coding. Type: leaderboard. Status: active.
Metric: Share of tasks solved (%) (higher is better).
Source: Scale AI Labs — https://labs.scale.com/leaderboard/swe_bench_pro_public_v2, results as of 2026-09-22, retrieved 2026-11-17, license: not stated.

| Rank | Model (scaffold, effort) | Score (%) |
|---|---|---|
| 1 | Opus 5 (Claude Code) xhigh | 98.00 |
| 2 | Fable 5.1 (Claude Code) high | 92.20 |
| 3 | Gemini 4 Argon (mini-swe-agent) high | 92.16 |
| 4 | GPT-6 Astra (Codex) high | 90.20 |
| 4 | Muse Wonder 1.0 (mini-swe-agent) xhigh | 90.20 |
| 6 | Kimi-K3 (mini-swe-agent) max | 88.20 |
| 6 | Sonnet 5 (Claude Code) xhigh | 88.20 |
| 8 | GPT-5.6 Terra (Codex) xhigh | 86.30 |
| 8 | Inkling 2 Small | 86.27 |
| 9 | Minimax M3p1 Flash (mini-swe-agent) xhigh | 84.31 |
| 10 | GLM-5.3 (mini-swe-agent) max | 84.30 |
| 11 | GPT-5.6 Sol (Codex) xhigh | 82.40 |
| 12 | Muse Spark Open (mini-swe-agent) xhigh | 82.35 |
| 13 | Sonnet 5.5 (mini-swe-agent) xhigh | 70.59 |
| 14 | Gemini 3.8 Flash (mini-swe-agent) high | 58.80 |
| 15 | Inkling (mini-swe-agent) xhigh | 56.90 |
| 16 | Haiku 5.5 (mini-swe-agent) high | 52.94 |
| 17 | Haiku 4.5 (Claude Code) xhigh | 25.50 |

Notes: SWE-Bench Pro V2 public split (642 tasks). Page has Full/HARD tabs; the table scraped is the default view and which tab it is was not identifiable from static HTML. Some entries gray-out/capped runs.
