Kimi K2.7 Code

Creator

Kimi

Released

2026-06-12

Intelligence

41.9

Artificial Analysis Index

Coding

60.8

Artificial Analysis Index

In $/1M

$0.95

input tokens

Out $/1M

$4.00

output tokens

Blended $/1M

$1.71

3:1 blended

Speed

36

tokens / sec

Profile

License, openness, and modality — the governance layer.

License

Modified MIT License

non-commercial

Weights

Open

downloadable

Modalities

Image · Text · Video

Parameters

1T

total

Kimi K2.7 Code is Moonshot AI's coding-focused agentic model built on Kimi K2.6. It improves real-world long-horizon coding task completion, instruction following, and token efficiency while reducing thinking-token usage by approximately 30% versus Kimi K2.6. The model uses a native multimodal Mixture-of-Experts architecture with 1T total parameters, 32B activated parameters, a 256K-token context window, image and video input support, and forced thinking with preserve_thinking enabled for multi-turn coding agent workflows.

Capability profile

Category strength across 20 domains, via LLM Stats.

tool calling
80math
70general
60
reasoning
60code
50agents
50
Via LLM Stats

Benchmark breakdown

Independent evaluation scores, normalized to 0–100.

GPQA Diamond
90
Humanity’s Last Exam
33
MMLU-Pro
SciCode
48
LiveCodeBench
MATH-500
AIME 2025
τ²-Bench (agentic)
90
Terminal-Bench Hard
45
IFBench
63
Via Artificial Analysis

Providers

Inference hosts serving this model — their own pricing and measured performance, cheapest input first.

ProviderIn $/1MOut $/1MThroughputLatencyStatus
DeepInfra$0.74$3.50active
Fireworks$0.95$4.00active
Moonshot AI$0.95$4.00active
Novita$0.95$4.0036active
Together$0.95$4.00active
All providers

5 hosts · via LLM Stats

© 2026 NYSGPT2525 LLC