SWE-Bench Pro leaderboard: who is #1

Name: SWE-Bench Pro rankings
Creator: Scale AI
License: https://creativecommons.org/licenses/by/4.0/

Scale AI · Resolved, higher is better · 11 models · published Sep 24, 2026

Scale AI

Top 11 of 11 models, best variant per model as published

#	Model	Score
1	Claude Opus 5as “Opus 5 (Claude Code) xhigh”	98.0%
2	Claude Fable 5.1as “Fable 5.1 (Claude Code) high”	92.2%
3	GPT-6 Astraas “GPT-6 Astra (Codex) high”	90.2%
4	Claude Sonnet 5as “Sonnet 5 (Claude Code) xhigh”	88.2%
4	Kimi K3as “Kimi-K3 (mini-swe-agent) max”	88.2%
6	GPT-5.6 Terraas “GPT-5.6 Terra (Codex) xhigh”	86.3%
7	GLM-5.3as “GLM-5.3 (mini-swe-agent) max”	84.3%
8	GPT-5.6 Solas “GPT-5.6 Sol (Codex) xhigh”	82.4%
9	Gemini 3.8 Flashas “Gemini 3.8 Flash (mini-swe-agent) high”	58.8%
10	Inklingas “Inkling (mini-swe-agent) xhigh”	56.9%
11	Claude 4.5 Haikuas “Haiku 4.5 (Claude Code) xhigh”	25.5%

About this board and its history

What it measures

Long-horizon software engineering tasks from real repositories: the public V2 set (relaunched September 2026), resolve rate on the Hard split the board opens on.

Other coding boards

Source

Scores as published by Scale AI on the capture date (2026-09-24). Source: labs.scale.com leaderboard page.

labs.scale.com/leaderboard/swe_bench_pro_public_v2

Sources: each leaderboard's own published results page, captured every day at 04:30 UTC. Scores exactly as each board published them; we run no evaluations. Logos via logo.dev; trademarks belong to their owners.

What it measures

Other coding boards

Source

New #1s by email