kimi · reasoning
KIMI

kimi-k3

Moonshot Kimi K3 (Coding API, 2.8T params MoE, 1M ctx, max reasoning) — #4 on Artificial Analysis intelligence index, subscription-billed via KIMI_API_KEY

⚖ COMPARE MODELS
VALS INDEX
Accuracy
VALS INDEX
17.6s
Latency (median)
$VALS INDEX
Cost / Test (1P-est)
SCORE STATUSnot scored
TEST SETS13
RECORDED TRIALS58
CUSTOM PASS78%

Industry benchmarks

Accuracyattempts · responded
GPQA_DIAMOND GPQA Diamond
not scored
10 attempts · 0 responded
HUMANEVAL_PLUS HumanEval+
not scored
10 attempts · 0 responded
IFEVAL IFEval
not scored
10 attempts · 0 responded
MMLU_PRO MMLU-Pro
not scored
10 attempts · 0 responded
Industry benchmark Bars = accuracy %, black cap = that score

Custom 9-task suite

Mean /5n=trials
CUSTOM code_debug
5.00
2 trials
CUSTOM code_gen_long
5.00
2 trials
CUSTOM json_strict
5.00
2 trials
CUSTOM reasoning_multistep
5.00
2 trials
CUSTOM agentic_tool_use
4.50
2 trials
CUSTOM summarize
4.50
2 trials
CUSTOM refactor_existing_code -
4.00
2 trials
CUSTOM agentic_prompt -
0.00
2 trials
CUSTOM creative_write -
0.00
2 trials
Custom suite · scored 0–5 by consensus (m3 + Fable-5) ★ = task scored a 5 at least once · sorted by mean score

Per-task distribution

Score

All trials on this page were run through the harness. Industry suites use deterministic family-specific reference graders (no LLM judge). The custom suite is graded by consensus of minimax-m3 + claude-fable-5. See /industry/ for the full benchmark suite.