ModelBeat
Models/Kimi K2.6
All models
M
moonshotai/kimi-k2.6

Kimi K2.6 is Moonshot’s open-weights AI model, released Apr 20, 2026. It costs $0.95/M input and $4/M output tokens, and handles a 262K-token context window. Its strongest use case is math, where it ranks in the 69th percentile of the models Model Beat tracks.

Kimi K2.6 is Moonshot AI's next-generation multimodal model, designed for long-horizon coding, coding-driven UI/UX generation, and multi-agent orchestration.

Context
262K
Input
$0.95 / 1M
Output
$4 / 1M
Cheapest via DigitalOcean
$0.57 / $2.40 / 1M
Released
Apr 20, 2026
Modalities
textimagetext
Strongest atMath top 31%Coding top 36%Agentic & Tools top 38%

Benchmarks

Scores on standardized evaluations. Higher is better — and rank shows where Kimi K2.6 lands among all models tracked on Model Beat.

63.0
Intelligence Index
Epoch AI
63rd percentile of tracked models
64.0
Coding Index
Epoch AI
64th percentile of tracked models
62.0
Agentic Index
Epoch AI
62nd percentile of tracked models

Reasoning

4 evals
GPQA Diamond90.8%

Graduate-level scientific reasoning

Humanity's Last Exam37.5%

Frontier of human expert knowledge

SimpleQA Verified34.9%

Factual accuracy & hallucination

WeirdML55.9%

Novel ML problem-solving

Coding

3 evals
SciCode51.5%

Scientific research coding

SWE-bench Verified76.7%

Real GitHub issue resolution

WebDev Arena1509

Human-rated web development

Math

3 evals
AIME 2024/202596.1%

Olympiad-qualifier math

FrontierMath39.0%

Research-level math problems

FrontierMath Tier 414.6%

Hardest research math

Agentic & Tools

2 evals
APEX18.9%

Multi-step agentic tasks

τ²-bench95.9%

Tool-agent-user reliability

In the news

12

Changelog

11
  • Kimi K2.6: cheapest credible provider via DigitalOcean now $0.57 per 1M input tokens (was $0.7, -19%)cheapest provider · Sep 17, 2026
  • Kimi K2.6: cheapest credible provider via DigitalOcean now $2.4 per 1M output tokens (was $3.5, -31%)cheapest provider · Sep 17, 2026
  • Kimi K2.6: reported SimpleQA Verified score fell from 38.7% to 34.9% (-10%)benchmark · Aug 28, 2026
  • Kimi K2.6: cheapest credible provider via Crusoe now $3.5 per 1M output tokens (was $3.2, +9%)cheapest provider · Aug 9, 2026
  • Kimi K2.6: cheapest credible provider via Crusoe now $0.7 per 1M input tokens (was $0.76, -8%)cheapest provider · Aug 9, 2026
  • Kimi K2.6: cheapest credible provider via DigitalOcean now $3.2 per 1M output tokens (was $3.5, -9%)cheapest provider · Aug 6, 2026
  • Kimi K2.6: cheapest credible provider via DigitalOcean now $0.76 per 1M input tokens (was $0.7, +9%)cheapest provider · Aug 6, 2026
  • Kimi K2.6: cheapest credible provider via Crusoe now $3.5 per 1M output tokens (was $3.2, +9%)cheapest provider · Jul 23, 2026
  • Kimi K2.6: cheapest credible provider via Crusoe now $0.7 per 1M input tokens (was $0.76, -8%)cheapest provider · Jul 23, 2026
  • Kimi K2.6: cheapest credible provider via DigitalOcean now $0.76 per 1M input tokens (was $0.8, -5%)cheapest provider · Jul 14, 2026
  • Kimi K2.6: cheapest credible provider via DigitalOcean now $3.2 per 1M output tokens (was $3.4, -6%)cheapest provider · Jul 14, 2026

Compare Kimi K2.6 with

Frequently asked questions

What is Kimi K2.6?

Kimi K2.6 is Moonshot AI's next-generation multimodal model, designed for long-horizon coding, coding-driven UI/UX generation, and multi-agent orchestration.

Who created Kimi K2.6?

Kimi K2.6 is an AI model developed by Moonshot, part of the Kimi family, released Apr 20, 2026.

How much does Kimi K2.6 cost?

Kimi K2.6's list price from Moonshot is $0.95 per million input tokens and $4 per million output tokens. The cheapest credible third-party provider on OpenRouter (DigitalOcean) serves it at $0.57/$2.40 per 1M.

What is Kimi K2.6's context window?

Kimi K2.6 supports a context window of up to 262K tokens.

How does Kimi K2.6 perform on benchmarks?

On standardized evaluations tracked by Epoch AI, Kimi K2.6 scores 95.9% on τ²-bench, 96.1% on AIME 2024/2025, 90.8% on GPQA Diamond.

Is Kimi K2.6 open source?

Kimi K2.6 is released with open weights (Open weights (unrestricted)), so the model can be downloaded and self-hosted.

Benchmark data from Epoch AI (CC BY) and Artificial Analysis; pricing, specs & descriptions from OpenRouter. Data updated Sep 17, 2026.