Kimi K3

Where people rank it

From blind votes on LMArena, where people pick the better of two anonymous answers. See the full tables on the AI leaderboard.

On paper

Fixed-test rankings from the benchmarks tab: Value #8, Reasoning #11, Speed #20.

Strengths and weaknesses

Best for

• The strongest open model for chat and coding
• Long documents
• Creative writing with an open model

Every score we have

Each bar compares this model with the best result any model on this page has on the same test.

• LMArena text arena (human votes)1,487.1Arena score 1487.1 (95% CI 1482.5 to 1491.7); 32,795 votes; entry "Kimi K3 (Max)"; position 18 of 416 on the full board
• LMArena creative writing1,461.3Arena score 1461.3 (95% CI 1453.1 to 1469.5); 6,646 votes; entry "Kimi K3 (Max)"; position 27 of 414 on the full board
• LMArena WebDev (coding)1,655.2Arena score 1655.2 (95% CI 1649.2 to 1661.2); 19,220 votes; entry "Kimi K3 (Max)"; position 14 of 144 on the full board
• LMArena Agent Arena0.019IPS score 0.0193 (95% CI 0.0120 to 0.0267); 135,415 sessions; entry "Kimi K3 (Max)"; position 23 of 52 on the full board
• Artificial Analysis Intelligence Index43.643.6 on the Artificial Analysis Intelligence Index v4.3.2 (AA shows 44), setting "Kimi K3 (max)"
• Humanity's Last Exam (Artificial Analysis)46.9%46.9% on Humanity's Last Exam (AA, 2,158 text-only questions), setting "Kimi K3 (max)"
• GPQA Diamond, graduate-level science93.5%93.5% on GPQA Diamond (AA; no longer part of the AA index since v4.2), setting "Kimi K3 (max)"
• Terminal-Bench 4.0 (Artificial Analysis run)12.6%12.6% on Terminal-Bench 4.0 as run by AA (mini-swe-agent harness), setting "Kimi K3 (max)"
• Long-context reasoning (AA-LCR)88.7%88.7% on AA-LCR (long-context reasoning), setting "Kimi K3 (max)"
• MMMU-Pro, image understanding80.5%80.5% on MMMU-Pro (image understanding, AA), setting "Kimi K3 (max)"
• AA-Omniscience, accuracy minus hallucination19.719.7 on the AA-Omniscience Index (knowledge accuracy minus hallucination, range -100 to 100), setting "Kimi K3 (max)"
• Output speed, tokens per second42 t/s42 output tokens per second (AA median), row "Kimi K3 (max)"
• Cost per Artificial Analysis test task$2.00$2.00 to run one AA Intelligence Index task, row "Kimi K3 (max)"
• Artificial Analysis Coding Agent Index51.951.9 on the AA Coding Agent Index v1.5 (DeepSWE v1.1, Terminal-Bench 4.0, SWE-Atlas-QnA), best listed setup "Kimi Code CLI - Kimi K3"
• ARC-AGI-2 puzzles60.4%60.42% on ARC-AGI-2 (semi-private set), best listed setting "Kimi K3 (Max)", $1.59 per task