Home Model series Gemini 4 Carbon Gemini 4 Carbon Gemini 4 Carbon · SWE-sweep · Version not stated in October 2026 table · Scores and configurations Gemini 4 Carbon · SWE-sweep · Version not stated in October 2026 table · Scores and configurations Gemini 4 Carbon record for SWE-sweep · Version not stated in October 2026 table. 1 score record for this release, including 1 exact value (may include historical records). Includes 1 source location.
Scores and evaluation conditions Gemini 4 Carbon · SWE-sweep scores and configurations Metric and score Configuration and evaluation conditions Sources Subtask-weighted reward 9.6% Reported 9.6% · Unit % Published Unconfirmed
Gemini 4 Carbon highest thinking Settings Gemini API model gemini-4-carbon; highest thinking settings; default sampling unless benchmark method states otherwise; pass@1 except where stated. Unknowns Exact thinking budget and sampling parameters are not published.、原文仅称 highest thinking settings,未确认对应的公开档位名称。 Tools Tool access not recorded Reasoning budget not recorded Framework mini-swe-agent Protocol Carbon self-computed with mini-swe-agent; subtask-weighted reward. Conditions Gemini 4 highest thinking; pass@1; version not stated. View this score in rankings
Benchmark versions and metrics Aliases SWE-sweep
Subtask-weighted reward · % · Higher is better