SWE-bench Verified (default "Bash Only" view, agent = mini-SWE-agent)

SWE-bench Verified (default "Bash Only" view, agent = mini-SWE-agent) is run by SWE-bench. It has ranked 47 models, of which the catalogue holds 24, scoring from 60 to 76.8 on % Resolved.

measured by SWE-bench · the board itself

Place	Model	Metric	Score
1stof 47	Claude Opus 4.5	% Resolved (high)	76.8
2ndof 47	Gemini 3 Flash Preview	% Resolved (high)	75.8
3rdof 47	MiniMax M2.5	% Resolved (high)	75.8
4thof 47	Claude Opus 4.6	% Resolved	75.6
5thof 47	Claude Opus 4.5	% Resolved (medium)	74.4
6thof 47	gemini-3-pro-preview	% Resolved	74.2
7thof 47	GPT-5.2 Codex	% Resolved	72.8
8thof 47	GLM-5	% Resolved (high)	72.8
9thof 47	GPT-5.2	% Resolved	72.8
9thof 47	GPT-5.2	% Resolved (high)	72.8
11thof 47	Claude Sonnet 4.5	% Resolved (high)	71.4
11thof 47	Claude Sonnet 4.5	% Resolved	71.4
12thof 47	Kimi K2.5	% Resolved (high)	70.8
14thof 47	DeepSeek v3.2	% Resolved (high)	70
15thof 47	gemini-3-pro-preview	% Resolved (high)	69.6
17thof 47	Claude Opus 4	% Resolved	67.6
18thof 47	Claude Haiku 4.5	% Resolved (high)	66.6
19thof 47	GPT-5.1 Codex	% Resolved (medium)	66
20thof 47	GPT-5.1	% Resolved (medium)	66
21stof 47	GPT-5	% Resolved (medium)	65
22ndof 47	Claude Sonnet 4	% Resolved	64.93
23rdof 47	Kimi K2 Thinking	% Resolved	63.4
24thof 47	MiniMax-M2	% Resolved	61
25thof 47	DeepSeek v3.2	% Resolved (reasoner)	60

Place

Model

Metric

Score

1stof 47

Claude Opus 4.5

% Resolved (high)

76.8

2ndof 47

Gemini 3 Flash Preview

% Resolved (high)

75.8