evals.report
BenchmarksLabsCompareRun guidesIn the wild

Kimi K3

Moonshot AI · Kimi. Released Jul 16, 2026.

Kimi K3 is a model from Moonshot AI in the Kimi family, released Jul 16, 2026. evals.report tracks 8 reported Kimi K3 benchmark scores across FrontierCode, AA-Omniscience: Knowledge and Hallucination Benchmark, ARC-AGI-1, ARC-AGI-2, DeepSWE, Terminal-Bench 2.1, SWE-Marathon, BrowseComp — each shown with its benchmark, metric, source status, and date, and never combined into a single ranking.

Open8 results

Benchmark results 8

Compare this model
BenchmarkCategoryScoreMetricStatusDate
FrontierCodeCoding44.2%weighted score (Main)OfficialJul 16, 2026Details
AA-Omniscience: Knowledge and Hallucination BenchmarkReasoning20AA-Omniscience IndexOfficialJul 16, 2026Details
ARC-AGI-1Reasoning94.5%accuracyOfficialJul 16, 2026Details
ARC-AGI-2Reasoning60.42%accuracyOfficialJul 16, 2026Details
DeepSWECoding67.5%% resolvedVerifiedJul 16, 2026Details
Terminal-Bench 2.1Agents88.3%task successVerifiedJul 16, 2026Details
SWE-MarathonAgents42.0%resolution rate (pass@1)VerifiedJul 16, 2026Details
BrowseCompAgents91.2%accuracyVerifiedJul 16, 2026Details