Kimi K2.6

Apr 2026
Kimi K2.6
Compare in
on
$0.71per task?
$0.95 / $4per 1M in / outcache $0.16
98 tok/sfirst token in 2.6 s?
256Kcontext

Rankingin thinking mode?

Full ranking →
1Sonnet 5.51,811.82Opus 5.51,799.73Muse Wonder Full1,729.34Fable 5.11,669.45Muse Wonder1,658.7
···
75Inkling833.176Kimi K2.6819.177Qwen3.6 27B815.4

Strong at

S
Tool calling#12 of 242Science#27 of 312Instructions#15 of 245
A
Reasoning#57 of 336Long context#36 of 314Knowledge#51 of 320Vision#24 of 125
B
Coding#65 of 134Agents & work tasks#66 of 177

Alternatives?

ModelScoreCostPer taskWhat changes
CheaperClaude Haiku 5.5maxwins 99%-70%$0.21Remove sampling parameters+5BetterGPT-6 Lunanon-thinkingwins 62%-98%$0.011Function calling only with none · Chat Completions+3OpenQwen3.8 Flashthinkingwins 99%-47%$0.37max_tokens range · QwenCloud+5CheaperDeepSeek V4.1 Flash 1017maxOpenwins 99%-63%$0.26Pass back reasoning_content · Chat Completions+4CheaperMiMo v2.6 ProthinkingOpenwins 98%-81%$0.13Pass back reasoning_content+3

Run it yourself

estimate
Fits on✗ 8 GB (phone, laptop)✗ 16 GB✗ 24 GB GPU✗ 48 GB GPU✗ 80 GB GPU
Licencemodified-mitcommercial use with conditions

History

5 versions
8001,0001,2001,4001,600Jul 26Release dateAA-Briefcase 1.1K2.6K3
8001,0001,2001,4001,600Jul 26Release dateAA-Briefcase 1.1K2.6K3

Benchmarks

in Agents & Tool Use
Terminal-Bench 2.165.9#42
Terminal-Bench 4.00.5#65
Terminal-Bench Hard (version unstated)43.9#22
GDPval-AA (work tasks) 2.11,038.4#66
AA-LCR (long context) 1.181#36
τ²-bench · AA95.9#12
τ³-Banking · AA23.3#45
APEX-Agents28.5#12
ITBench SRE31.2#28
AutomationBench-AA13#75
EnterpriseOps-Gym-AA38.5#24
AA-Briefcase 1.1 · Elo819.1#76
HLE without toolslab36.4
SWE-bench Pro (edition unstated)lab58.6
BrowseComplab83.2
HLE with toolslab54
MCPAtlas Publiclab66.6
Toolathlonlab50
B3 Attack Resistancelab83.8
Harmful cyber request refusal meanlab90.7
KORA aggregatelab1.6

FAQ

How much does Kimi K2.6 cost?›

$0.95 per 1M input tokens and $4 per 1M output tokens; cached input $0.16 per 1M; cheapest via Chutes at $0.45 / $1.95. A typical benchmark task costs about $0.71 at default effort.

Is Kimi K2.6 free to use?›

There is no free API; the cheapest paid price is $0.45 per 1M input tokens. The weights are open, so you can run it yourself.

How good is Kimi K2.6, and what is it best at?›

#57 of 336 in Chat & Knowledge, #65 of 134 in Coding, #75 of 135 in Agents & Tool Use, #40 of 197 in Writing & Language, #24 of 125 in Vision & Documents. Strongest at: Tool calling, Reasoning, Science, Long context, Knowledge, Instructions, Vision.

Is Kimi K2.6 better than Inkling?›

On AA-Briefcase 1.1 · Elo, Kimi K2.6 scores 819.1 vs 833.1 — about the same. It is 19% more expensive per task.

How can I use Kimi K2.6?›

Through the API of AtlasCloud, Azure AI Foundry, Chutes, Cloudflare, CoreWeave and 9 more.

What is the context window of Kimi K2.6?›

256K tokens, with up to 236K output tokens.

When was Kimi K2.6 released? Is there a newer version?›

Released 2026-04-20. It has been superseded by Kimi K3.

Is Kimi K2.6 open source?›

Yes, the weights are open (modified-mit licence).

Can I run Kimi K2.6 locally?›

The weights are open and can be self-hosted; hardware needs are not confirmed yet.

Can Kimi K2.6 read images?›

It accepts text and images as input.

What is Kimi K2.6?›

Kimi K2.6 is an open-weights language model from Moonshot AI, released Apr 2026, the successor to Kimi K2.5.

What are the best alternatives to Kimi K2.6?›

Claude Haiku 5.5 (cheaper), GPT-6 Luna (better), Qwen3.8 Flash (open weights).