SWE-bench Verified (Bash Only toggle off — all agents)

SWE-bench Verified (Bash Only toggle off — all agents) is run by SWE-bench. It has ranked 180 models, of which the catalogue holds 13, scoring from 72.8 to 79.2 on % Resolved.

measured by SWE-bench · the board itself

Place	Model	Metric	Score
1stof 180	Claude Opus 4.5	% Resolved (medium)	79.2
3rdof 180	Doubao-Seed-code	% Resolved	78.8
4thof 180	gemini-3-pro-preview	% Resolved	77.4
5thof 180	Rovo Dev	% Resolved	76.8
6thof 180	Claude Sonnet 4	% Resolved	76.8
7thof 180	Claude Opus 4.5	% Resolved (high)	76.8
9thof 180	Gemini 3 Flash Preview	% Resolved (high)	75.8
10thof 180	MiniMax M2.5	% Resolved (high)	75.8
12thof 180	Claude Opus 4.6	% Resolved	75.6
15thof 180	Claude Sonnet 4.5	% Resolved	74.8
19thof 180	GPT-5	% Resolved	74.4
23rdof 180	Claude Opus 4	% Resolved	73.2
25thof 180	GPT-5.2 Codex	% Resolved	72.8

Place

Model

Metric

Score

1stof 180

Claude Opus 4.5

% Resolved (medium)

79.2

3rdof 180

Doubao-Seed-code

% Resolved

78.8