What it is
Published Best in Agentic Coding (SWE Bench) scores.
← Back to leaderboards
BENCHMARK RECORD
SWE-Bench Verified
Source-linked result details from the TopLLM database.
BENCHMARK RECORD · 2026-10-11
SWE-Bench Verified
What it tests
Published Best in Agentic Coding (SWE Bench) scores.
How to read it
Scores are benchmark-specific and missing values are not estimated.
10 published entries96.2 top reported scorecoding evaluation domain
Published leaders
Each bar is scaled only within this benchmark.
01GPT-5.6 Sol96.202Claude Mythos 595.5
03Claude Fable 595.0Unresolved model record
04GPT-5.6 Luna93.0Unresolved model record
05Claude Opus 4.888.6Unresolved model record
06DeepSeek V4 Pro80.607MiniMax M380.508Kimi K2.680.209DeepSeek V4 Flash79.010Kimi K2.576.8Scores are benchmark-specific. Missing values remain N/A and are never inferred from another evaluation.
Compare in the benchmark matrix →