Benchmark · KnowledgeCompetitive

LiveBench · Agentic Coding

Updated 2026-06-16
Models tested
40
Top score
73.3
GLM 5.2
Median
45.0
min 3.3
Top-5 spread
σ 7.0
wide open

Best score over time · one chart, every benchmark

LIVEBENCH · AGENTIC CODING40 MODELS · FRONTIER RUNNING MAX0255075100SCORE ↑Jul 25Oct 25Jan 26Mar 26Jun 26RELEASE DATE →benchgecko.ai/benchmark/livebench-agentic-coding · frontier
Frontier on LiveBench · Agentic Coding rose from 13.3 to 73.3 in 11 months · +60.0 points · latest leader GLM 5.2 from z-ai.
Pink dots = frontier records · 10 totalClick to open model page

What it measures · how it is scored · sources · as of 2026-11-18

  • What it measures · LiveBench coding tasks that require multi-step reasoning and tool use. Tests planning and execution of complex coding workflows.
  • How it is scored · Scored as a percentage · higher is better · maximum 100
  • Scores from · LiveBench (40 scores)

Same category · related evaluations