Kimi K2.5

by Moonshot

Released 82 days after Kimi K2 Thinking · All Moonshot releases

Kimi K2.5, released January 27, 2026, is Moonshot's native multimodal agentic model, integrating vision and language through continual pretraining on roughly 15 trillion mixed visual and text tokens. It keeps the 1 trillion total parameter MoE architecture with 32 billion active parameters and adds visual understanding through a MoonViT vision encoder of about 400 million parameters. The model handles code generation from visual specifications, visual reasoning across UI designs and video, and tool orchestration grounded in visual inputs. It supports a 256K token context window and offers several operational modes ranging from instant responses to deliberate reasoning and multi-agent coordination. A standout capability is agent swarm coordination, which can coordinate up to 100 parallel sub-agents and around 1,500 tool calls per task to decompose and execute complex work, with roughly 4.5 times faster execution than a single agent on wide tasks. It reports strong results on coding benchmarks, including 76.8% on SWE-Bench Verified.

Call Kimi K2.5 on Opper with the OpenAI SDK. Sign up without a credit card. Get started

Open in playgroundTry in chatCompare

Key info

Input
Output
Features
Context window
262K
Max output
16K
Input price
$0.60 /1M$0.12 cache read
Output price
$3.13 /1M
Released
  • EU/EEA residency available
  • US residency available
  • Zero data retention on pay-as-you-go
  • No training by default
  • GDPR DPA available

Available routes

Kimi K2.5 runs on 7 different routes through the Opper gateway. Compare residency, zero data retention and training posture at a glance, with full data-handling detail per route below.

ProviderRegionZero data retentionTrainingInputOutputCache read
Alibaba Cloud
MultiNo$0.60$3.13$0.12
DeepInfra
USNo$0.45$2.25$0.07
DigitalOcean
USNo$0.50$2.70$0.20
Geodd
EU/EEANo$0.40$2.60Input price
Geodd
USNo$0.40$2.60Input price
Novita
USNo$0.60$3.00$0.10
TensorX
EUNo$0.50$2.80$0.13

Uptime and availability

Kimi K2.5 runs on 3 independently monitored routes through the Opper gateway. A direct integration leaves you on one of them, the gateway serves whichever is healthy.

100%effective uptime, last 30 days
Averaged across these routes on their own, a single provider reached 100% over the same window.
Daily status through the gateway, which serves each request from whichever route is healthy.
3 monitored providers serve Kimi K2.5.

Measured over the last 30 days from each provider's official status feed via StatusGator. Refreshed hourly. 3 further routes for this model are not on the board yet and are not counted above. See uptime for every provider Opper monitors.

Data handling per route

Each route hosting Kimi K2.5 has its own privacy posture, residency, and GDPR terms. Postures are maintained by Opper with a last-verification timestamp.

Alibaba Cloud — Multi-region

Zero data retention with abuse monitoring: prompts and outputs are kept for abuse monitoring only, for a window the provider does not state, and are not used for training. No training on customer data. GLOBAL; SCCs; DPA available.

Zero data retention
With abuse monitoring, window not stated.
Training
No training on customer data.
Logging
Abuse monitoring
Abuse monitoring
On by default, holds flagged content
Caching
Content cached for replay
GDPR DPA
DPA available
Transfer mechanism
SCCs

DeepInfra — United States🇺🇸

Zero data retention: nothing is logged, held for abuse monitoring or used for training. No training on customer data. US; unknown; DPA available.

Zero data retention
Yes. Nothing is logged, held for abuse monitoring or used for training.
Training
No training on customer data.
Logging
None
Abuse monitoring
No classifier
GDPR DPA
DPA available
Transfer mechanism
unknown

DigitalOcean — United States🇺🇸

Zero data retention with abuse monitoring: nothing is logged or used for training; content flagged by abuse monitoring is held for review. No training on customer data. US; dpf sccs; DPA available.

Zero data retention
With abuse monitoring, flagged content held.
Training
No training on customer data.
Logging
None
Abuse monitoring
Always on, holds flagged content, stored in United States
Caching
Content cached for replay
Subprocessor access
No subprocessor reads content
GDPR DPA
DPA available
Transfer mechanism
dpf sccs

Geodd — Norway🇳🇴

Zero data retention: nothing is logged, held for abuse monitoring or used for training. No training on customer data. EEA; SCCs; DPA available.

Zero data retention
Yes. Nothing is logged, held for abuse monitoring or used for training.
Training
No training on customer data.
Logging
None
Abuse monitoring
No classifier
Caching
Not cached
GDPR DPA
DPA available
Transfer mechanism
SCCs

Geodd — United States🇺🇸

Zero data retention: nothing is logged, held for abuse monitoring or used for training. No training on customer data. US; SCCs; DPA available.

Zero data retention
Yes. Nothing is logged, held for abuse monitoring or used for training.
Training
No training on customer data.
Logging
None
Abuse monitoring
No classifier
Caching
Not cached
GDPR DPA
DPA available
Transfer mechanism
SCCs

Novita — United States🇺🇸

Zero data retention: nothing is logged, held for abuse monitoring or used for training. No training on customer data. US; SCCs; DPA available.

Zero data retention
Yes. Nothing is logged, held for abuse monitoring or used for training.
Training
No training on customer data.
Logging
None
Abuse monitoring
On by default, keeps nothing
GDPR DPA
DPA available
Transfer mechanism
SCCs

TensorX — European Union🇪🇺

Zero data retention: nothing is logged, held for abuse monitoring or used for training. No training on customer data. EU; DPA available.

Zero data retention
Yes. Nothing is logged, held for abuse monitoring or used for training.
Training
No training on customer data.
Logging
None
Abuse monitoring
No classifier
Caching
Content cached for replay
GDPR DPA
DPA available
Transfer mechanism
Not applicable — data stays in EU

Benchmarks

Independent benchmark scores — composite indices for reasoning, coding, and math, plus individual eval scores where available.

Global rank#164 of 665 LLMs
TierEfficient
Intelligence Index23.5
Coding Index46.8
Reasoning & knowledge
GPQA Diamond
88%
Humanity's Last Exam
31%
Long-context reasoning
78%
Agentic & tool use
Terminal-Bench Hard
35%
τ²-Bench Telecom
96%
Math & instruction following
IFBench
70%
See full leaderboard →Benchmarks via Artificial Analysis · View on AA

Get started

Call Kimi K2.5 through the Opper gateway with one API key. Opper is drop-in compatible with the OpenAI, Anthropic and Google AI SDKs.

Kimi K2.5 is a premium model on Opper. Sign up needs no credit card: you get an API key straight away and the free models work in the playground and the API. Add a card to use premium models, pay-as-you-go with no minimum.

Set it up with your agent

Copy this and paste it into a coding agent like Claude Code, Cursor or Codex and it'll wire up Opper for you.

Or call it directly

import OpenAI from "openai";
const client = new OpenAI({
  apiKey: process.env.OPPER_API_KEY,
  baseURL: "https://api.opper.ai/v3/compat",
});
const completion = await client.chat.completions.create({
  model: "kimi-k2.5",
  messages: [{ role: "user", content: "Hello" }],
});
console.log(completion.choices[0].message.content);

Compare Kimi K2.5 with…

Side-by-side on privacy, EU hosting, pricing, and benchmarks.

Kimi K2.5 vs Claude Opus 5Kimi K2.5 vs GPT-5.4Kimi K2.5 vs Gemini 3.5 FlashKimi K2.5 vs Kimi K2.6Kimi K2.5 vs Kimi K3Build your own comparison

Other models from Moonshot

Kimi K3
Moonshot
Kimi K2.6
Moonshot
Kimi K2.7 Code
Moonshot
Kimi K2 0905
Moonshot
Kimi K2 Instruct
Moonshot

Start building with 700+ models

One API key. Every major provider. Up and running in minutes.

Get startedView Documentation