Model

Kimi k1.5

CLOSED

Moonshot AI · Kimi k1.5 family · released Jan 20, 2025

Moonshot's first RL-trained multimodal reasoning model, claiming OpenAI o1-level math, coding, and vision performance.

ReasoningCodingVisionFunction callingTool useAgentic

1391.2

Elo · rank #138

Parameters

Undisclosed

Active params

—

Context

128K tokens

Architecture

Transformer, RL-trained long/short-CoT reasoning (parameter count undisclosed)

License

Proprietary

Languages

—

API price (in/out)

No hosted API

Modalities

text · vision

Benchmark results

Bar shows position within the tracked field; marker = field best

AIMEMath77.5%#86

best: GPT-5.2 · 100.0%

C-EvalKnowledge88.3%#16

best: Qwen3.6-Plus · 93.3%

CodeforcesCoding94#45

best: DeepSeek-V4-Pro (Think Max) · 3206

IFEvalReasoning87.2%#44

best: Gemma 4 26B A4B · 98.5%

LiveCodeBenchCoding47.3%#122

best: DeepSeek-V4-Pro (Think Max) · 93.5%

MATH-500Math96.2%#27

best: GPT-5 · 99.4%

MathVisionVision38.6%#15

best: Seed 2.1 Pro · 92.6%

MathVistaVision74.9%#16

best: Seed 2.1 Pro · 90.7%

MMLUKnowledge87.4%#33

best: OpenAI o3 · 92.9%

MMMUVision70.0%#56

best: Claude Fable 5 · 89.3%

Run it locally

Closed weights — available via API only. No local deployment.

Input / M tok

—

Output / M tok

—

Kimi k1.5 family

Elo progression across releases

Kimi k1.5Jan 20251391.2

Resources

GHMoonshotAI/Kimi-k1.5 DOCSarxiv.org/abs/2501.12599

API price — · each benchmark row carries its own source badge (see methodology)