Skip to content
Leaders Board.
Models/Moonshot AIOpen weights

Kimi K3

#18 of all models across the language-model boards.
#18
Overall rank
9.3
Podium score
0
Board wins
Jul 16, 2026
Released
·Chat·Reasoning4Coding·Agents and tools·Speed·Price·Usage

Chat

Which model people prefer in blind head-to-head conversations.

14
Text Arena
1,485 · of 100
as “kimi-k3-max”

Reasoning

Hard questions with checkable answers: science, math, puzzles, expert exams.

12
AA Intelligence
43.6 · of 100
as “Kimi K3 (max)”
12
Epoch ECI
157.7 · of 100
as “Kimi K3”
12
LiveBench
79.2% · of 59
as “kimi-k3”
15
GPQA Diamond
93.1% · of 100
as “kimi-k3_max”
25
ARC-AGI-2
60.4% · of 78
as “Kimi K3 (Max)”

Coding

Fixing real repositories, building web apps, editing code.

4
SWE-Bench Pro
88.2% · of 11
as “Kimi-K3 (mini-swe-agent) max”
7
WebDev Arena
1,660 · of 100
as “kimi-k3-max”

Agents and tools

Calling functions and tools, using MCP servers, searching the web.

8
MCP Atlas
82.3% · of 34
as “kimi-k3 (max)”

Speed

Output tokens per second on the vendor API.

23
AA Speed
38 tok/s · of 24
as “Kimi K3 (low)”

Price

Blended API price per million tokens, cheapest first.

24
AA Price
$6.00 · of 27
as “Kimi K3 (max)”

Usage

What developers actually send tokens to.

18
OpenRouter
1.4T tok · of 19
as “moonshotai/kimi-k3-20260715”

More from Moonshot AI

Kimi K2.5Moonshotai Kimi K2Kimi K2.7 CodeKimi K2.6Kimi K2P5Kimi K2 TurboKimi K2 ThinkingKimi K2.5 Instant

Scores as published by each board on the capture date. Model names and logos belong to their owners; logos via logo.dev.

Weekly: the AI leaderboards with a new #1, Saturday mornings.