Kimi K2.5

MoonshotAI· released 27 Jan 2026

moonshotai/kimi-k2.5
Input / 1M tokens
$0.45
Output / 1M tokens
$2.25
Context window262K tokens

Longest answer: 236K tokens

reads textreads imageopen weightstool callingvisionreasoning

About

Kimi K2.5 is a multimodal model from MoonshotAI, released on 2026-01-27. It accepts text and images and returns text, with a context window of 262,144 tokens. The model supports tools, vision and reasoning, and its maximum output is 235,929 tokens. MoonshotAI lists open weights, with Hugging Face ID moonshotai/Kimi-K2.5. Its model ID is moonshotai/kimi-k2.5. Input costs USD 0.45 per 1M tokens and output costs USD 2.25 per 1M tokens. Its summary describes visual coding capability and a self-directed agent swarm approach, with continued pretraining over approximately 15T mixed tokens. For builders considering image-aware coding or agentic work, the listed input types, tool support, context and rates are practical points to weigh. The official page is on OpenRouter.

Who it is for

Kimi K2.5 suits teams considering image-aware coding or agentic workflows, with tools, vision and reasoning support. Its text and image inputs may fit builders whose tasks include visual material.

What is good

  • Open weights are listed.
  • Accepts text and image inputs.
  • Supports tools, vision and reasoning.
  • Context window is 262,144 tokens.
  • Maximum output is 235,929 tokens.

What to know first

  • Outputs text only.
  • Output costs USD 2.25 per 1M tokens.
  • Released on 2026-01-27.

Verdict

Kimi K2.5 brings image input, tool support and reasoning to text-output tasks, with a stated focus on visual coding and agent swarms. Its listed output rate is higher than its input rate, a consideration for token-heavy workflows.

Details

Lab
MoonshotAIopenrouter.ai · 3 Oct 2026
Context
262Kopenrouter.ai · 3 Oct 2026
Input price
$0.45 / 1Mopenrouter.ai · 3 Oct 2026
Output price
$2.25 / 1Mopenrouter.ai · 3 Oct 2026
Max output
235,929 tokensopenrouter.ai · 3 Oct 2026
Inputs
text, imageopenrouter.ai · 3 Oct 2026
Open weights
Yesopenrouter.ai · 3 Oct 2026

More longest-context models

See the list