DataLearner logo
KI

Kimi K2.5

Multimodal modelKimi K2Kimi K2.5

Kimi K2.5

Release date: 2026-01-27Updated: 2026-07-17 21:57:43.155Knowledge cutoff: 2024-046,293
Live demoGitHubHugging FaceCompare
Parameters
1T
Context length
256K
Chinese support
Supported
Reasoning ability

Kimi K2.5 is a multimodal model from Moonshot AI, released on 2026-01-27. It accepts text and image input and returns text output. The cataloged parameter count is 1T, with 32B active parameters per inference. The recorded context window is 256K, and the recorded maximum output is 16K. Cataloged capabilities include Reasoning model and Multilingual. The checkpoint is listed under the Modified Mit License, with the license link included in the model references. Use the linked references to confirm current access, licensing, and provider-specific limits.

Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology

Kimi K2.5

Model basics

Reasoning traces
Supported
Thinking modes
Standard ModeThinking Level · Extended
Context length
256K tokens
Max output length
16K tokens
Model type
Multimodal model
Modality (in / out)
Text, Image → Text
Release date
2026-01-27
Model file size
595GB
MoE architecture
Yes
Total params / Active params
1T / 32B
Knowledge cutoff
2024-04
Kimi K2.5

Open source & experience

Kimi K2.5

Official resources

Paper
Paper
DataLearnerAI blog
DataLearnerAI blog
Kimi K2.5

API details

API speed
2/5
💡Default unit: $/1M tokens. If vendors use other units, follow their published pricing.
Learn about pricing modes
Standard
TypeConditionInputOutput
Text-$0.600/ 1M$3.00/ 1M
Image-$0.600/ 1M-
Cache PricingPrompt Cache
TypeTTLWriteRead
Text--$0.100/ 1M
Image--$0.100/ 1M
Kimi K2.5

Benchmark Results

Kimi K2.5 currently shows benchmark results led by LiveCodeBench (16 / 123, score 85), HLE (27 / 172, score 50.20), AIME2025 (21 / 107, score 96.10). This page also consolidates core specs, context limits, and API pricing so you can evaluate the model from benchmark results and deployment constraints together.

Thinking
Tool usage
Internet

General Knowledge

7 evaluations
Benchmark / mode
Score
Rank/total
GPQA Diamond
Thinking Mode
87.60
37 / 187
MMLU Pro
Thinking Mode
78.50
69 / 132
LiveBench
Thinking Mode
69.07
42 / 115
ARC-AGI
Thinking Mode
65.30
34 / 68
HLE
Thinking Mode
30.10
89 / 172
HLE
Thinking ModeTools
50.20
27 / 172
ARC-AGI-2
Thinking Mode
11.80
39 / 62

Coding and Software Engineer

4 evaluations
Benchmark / mode
Score
Rank/total
LiveCodeBench
Thinking Mode
85
16 / 123
SWE-bench Verified
Thinking ModeTools
76.80
30 / 112
SWE-bench Multilingual
Thinking Mode
73
13 / 23
SWE-Bench Pro - Public
Thinking ModeTools
50.70
41 / 54

Math and Reasoning

4 evaluations
Benchmark / mode
Score
Rank/total
AIME2025
Thinking Mode
96.10
21 / 107
AIME 2026
Thinking Mode
92.50
12 / 18
IMO-AnswerBench
Thinking Mode
81.80
16 / 21
FrontierMath - Tier 4
Standard Mode
4.20
40 / 80

Common Sense Reasoning

1 evaluations
Benchmark / mode
Score
Rank/total
Simple Bench
Thinking Mode
46.80
30 / 63

AI Agent - Information Search

1 evaluations
Benchmark / mode
Score
Rank/total
BrowseComp
Thinking ModeToolsInternet
60.60
36 / 53

AI Agent - Tool Usage

2 evaluations
Benchmark / mode
Score
Rank/total
MCP-Atlas
Standard ModeTools
64.40
19 / 27
Terminal Bench 2.0
Thinking ModeTools
50.80
34 / 47

Productivity Knowledge

1 evaluations
Benchmark / mode
Score
Rank/total
GDPval-AA
Thinking Mode
40
15 / 21

Long Context

2 evaluations
Benchmark / mode
Score
Rank/total
AA-LCR
Thinking Mode
65
12 / 15
LongBench v2
Standard Mode
61
5 / 11

Claw-style Agent Evaluation

2 evaluations
Benchmark / mode
Score
Rank/total
Pinch Bench
Thinking ModeTools
84.80
17 / 37
Claw Bench
Thinking ModeTools
81.70
18 / 29
View benchmark analysisCompare with other models

Compare with other models

Kimi K2.5

Publisher

Kimi K2.5

Model Overview

Kimi K2.5 is a multimodal model from Moonshot AI, released on 2026-01-27.

It accepts text and image input and produces text output. Its cataloged capabilities include Reasoning model and Multilingual. The cataloged parameter count is 1T, with 32B active parameters per inference. The recorded context window is 256K, and the recorded maximum output is 16K.

The checkpoint is listed under the Modified Mit License, with the license link included in the model references. The page records 5 API pricing rules from Moonshot AI; current provider pricing and conditions should be checked before deployment. The evaluation section contains 24 cataloged benchmark results with their recorded modes and scores. The page links 5 release, model-card, repository, or provider references for checking the underlying claims. Specifications, availability, and prices can change; undisclosed values are intentionally left unstated.

Kimi K2.5

FAQ

What is Kimi K2.5?

Kimi K2.5 is a multimodal model from Moonshot AI, released on 2026-01-27. It accepts text and image input and returns text output. The cataloged parameter count is 1T, with 32B active parameters per inference. The recorded context window is 256K, and the recorded maximum output is 16K. Cataloged capabilities include Reasoning model and Multilingual. The checkpoint is listed under the Modified Mit License, with the license link included in the model references. Use the linked references to confirm current access, licensing, and provider-specific limits.

What input and output modalities does Kimi K2.5 support?

The current model record lists text and image as input and text as output.

What are the main recorded specifications for Kimi K2.5?

The cataloged parameter count is 1T, with 32B active parameters per inference. The recorded context window is 256K, and the recorded maximum output is 16K. Fields without a source-backed value remain undisclosed.

Does Kimi K2.5 have API pricing?

The page records 5 API pricing rules from Moonshot AI; current provider pricing and conditions should be checked before deployment.

Are benchmark results available for Kimi K2.5?

The evaluation section contains 24 cataloged benchmark results with their recorded modes and scores. Compare only results that use the same benchmark version and evaluation mode.

Is Kimi K2.5 open source?

The checkpoint is listed under the Modified Mit License, with the license link included in the model references. Review the linked license text before commercial or derivative use.

DataLearner on WeChat

Follow DataLearner on WeChat for AI model updates and research notes.

DataLearner WeChat QR code