#02Moonshot· released 2026-07-16

MoonshotAI: Kimi K3.

Kimi K3 is a 2.8T parameter open-weight multimodal reasoning model from Moonshot AI. It is suited for complex coding, knowledge work, and long-horizon agentic workflows, and is particularly strong at...

View on OpenRouter ↗
Context
1.0M
Input $/1M
$3.00
Output $/1M
$15.00
Modality
text · image

Benchmarks

How these are aggregated →
Arena Elo
1571
LMArena — human blind votes
AA Index
70.0
Artificial Analysis composite
LiveBench
83.9
Contamination-resistant
GPQA
90.6%
Grad-level science reasoning
MMLU-Pro
88.1%
Broad knowledge
SWE-bench
76.8%
Real GitHub bugs (Verified)
Terminal-Bench
55.2%
Agentic shell tasks
ARC-AGI
21.8%
Abstract reasoning
Aider Polyglot
87.4%
Multi-language code editing
Speed
52 tok/s
Median output speed

Nearby on the leaderboard