System Card← leaderboard
Moonshot AI

Kimi

#3of 28 · general

Kimi K3.1 Pro

2.8Tparams · MoE
1Mcontext
$2 / $10per 1M in / out
2026-10-23released
50100Coding 97 · #1 of 28 · one sourceAgents 87 · #2 of 29 · one sourceCyber 16 · #16 of 24 · one sourceEngineering 71 · #8 of 26 · one sourceMath: no readingHealth: no readingLegal 61 · #8 of 27 · one sourceProfessional 72 · #6 of 28 · one sourceWriting: no readingInstruct: no readingHallucination: no readingMultimodal 57 · #9 of 24 · one source
Coding97#1
Agents87#2
Cyber16#16
Engineering71#8
Math—
Health—
Legal61#8
Professional72#6
Writing—
Instruct—
Hallucination—
Multimodal57#9

History

0%10%20%30%Jul 2025Oct 2025Jan 2026Apr 2026Jul 2026Oct 2026Kimi K2 · $0.57/M in · 2025-07-12 · still served · CritPt 0.0% fractional score (0–1 scale)Kimi K2 Thinking · $0.6/M in · 2025-11-06 · still served · CritPt 2.6% fractional score (0–1 scale)Kimi K2.5 · $0.49/M in · 2026-01-27 · still served · CritPt 3.1% fractional score (0–1 scale)Kimi K2.6 · $0.4323/M in · 2026-04-20 · still served · CritPt 8.0% fractional score (0–1 scale)Kimi K3 · $0.73/M in · 2026-07-16 · still served · CritPt 13.3% fractional score (0–1 scale)Kimi K3.1 Pro · $2/M in · 2026-10-23 · still served · CritPt 24.6% fractional score (0–1 scale)K3K2.6K2.5K2 ThinkingK2
  • Moonshot AI
  • older → newer
$ / M input tokens6 of 176 priced models · price is a third-party claim, not a vendor quote

Usage

1.2TtokensAug 2 – Oct 6▼ 19% 7d
400B800B1.2T1.6TAug 2Sep 1Oct 1

Geography

United States28.8%
China10.9%
Germany4.2%
Singapore3.8%
Brazil3.3%
Japan3.3%
Hong Kong3.1%
Spain2.9%
India2.3%
France2.3%

Measured

≈ imputed · amber dot = lab-self-reported among the sources
HLE52.5%#7Artificial Analysis
SciCode60.8%#7Artificial Analysis
CritPt24.6%#10Artificial Analysis
AA Index55.1#4Artificial Analysis

Neighbors

Claude Sonnet→ weaker codingGPT Astra→ stronger multimodal · pricierGPT Sol→ weaker agent · weaker codingClaude Opus→ stronger multimodal · pricierMuse Wonder→ weaker multimodal · pricier