evals.report
BenchmarksLabsCompareRun guidesIn the wild

Kimi K2 Instruct

Moonshot AI · Kimi. Released Jul 11, 2025.

Kimi K2 Instruct is a model from Moonshot AI in the Kimi family, released Jul 11, 2025. evals.report tracks 19 reported Kimi K2 Instruct benchmark scores across Berkeley Function Calling Leaderboard, SWE-bench Pro, WeirdML, MCP Atlas, Artificial Analysis Intelligence Index, Epoch Capabilities Index, Aider Polyglot, MMLU-Pro, and 11 more — each shown with its benchmark, metric, source status, and date, and never combined into a single ranking.

Open19 results

Benchmark results 19

Compare this model
BenchmarkCategoryScoreMetricStatusDate
Berkeley Function Calling LeaderboardTool use59.06%accuracyOfficialJul 11, 2025Details
SWE-bench ProCoding27.67%% resolvedOfficialJul 11, 2025Details
WeirdMLCoding39.4%average accuracyOfficialJul 11, 2025Details
MCP AtlasTool use23.9%pass rateOfficialJul 11, 2025Details
Artificial Analysis Intelligence IndexReasoning26.3IndexUnverifiedJul 11, 2025Details
Epoch Capabilities IndexReasoning141.2IndexOfficialJul 11, 2025Details
Aider PolyglotCoding60.0%% correctUnverifiedJul 11, 2025Details
MMLU-ProReasoning84.8%accuracyVerifiedJul 11, 2025Details
τ²-bench (Telecom)Tool use65.8%pass^1UnverifiedJul 11, 2025Details
LiveCodeBenchCoding85.3%Pass@1UnverifiedJul 11, 2025Details
SciCodeCoding42.4%accuracyUnverifiedJul 11, 2025Details
LongBench v2Reasoning44.3%accuracyOfficialJul 11, 2025Details
EQ-Bench Creative Writing v3Chat preference1738EloVerifiedJul 11, 2025Details
Design ArenaChat preference1088EloVerifiedJul 11, 2025Details
MASK (Model Alignment between Statements and Knowledge)Other46.67Honesty scoreVerifiedJul 11, 2025Details
SuperGPQAReasoning57.2%accuracyUnverifiedJul 11, 2025Details
IMO-BenchReasoning45.8%accuracyVerifiedJul 11, 2025Details
PolyMath: Evaluating Mathematical Reasoning in Multilingual ContextsReasoning65.1%Difficulty-Weighted Accuracy (DW-ACC)UnverifiedJul 11, 2025Details
GSO: Software Optimization Benchmark for SWE-AgentsCoding4.9%Opt@1OfficialJul 11, 2025Details