LLM Reference

Kimi Models by Moonshot AI

Moonshot AIProprietary
2 models2026Up to 1.05m ctxFrom $0.44/1M input

Last refreshed 2026-08-24. Next refresh: weekly.

Details

ResearcherMoonshot AI
LicenseProprietary
Commercial useCommercial use: conditional
Models2
Released2026
Max context1.05m

Capabilities

VisionAll models
MultimodalAll models
Reasoning1 of 2 models
JSON / Tool useAll models
Structured OutputsAll models

Links

Website

About

Kimi, created by Moonshot AI, is a pioneering family of large language models renowned for their ability to process extraordinarily long contexts. Initially supporting up to 200,000 Chinese characters (roughly 400 million tokens) 410, its capacity has soared to an impressive 2 million Chinese characters 346, outstripping many leading models like GPT-4's 128,000 tokens 4. This extensive context capability enables Kimi to efficiently summarize long documents, analyze intricate research papers, and carry out complex dialogues. Although primarily designed for the Chinese language, Kimi offers support for several other languages 1410, making it adaptable for applications ranging from scholarly research to software engineering 410. Its continuous evolution is a testament to the swift progress in AI technology 8.

Decision facts

Best fit
codevision and multimodal workreasoning
Capability starting point
Kimi K3 with 1.05m context and reasoning, JSON / Tool use, structured outputs, and multimodal inputs
Lowest tracked input
Kimi K2.5 · $0.44/1M · OpenRouter
Closest related family
Kimi K2

Current Variants

Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.

2 in view
Kimi K3Current

Use when the workload needs 1.05m context, reasoning, and JSON / Tool use.

2026-071.05m contextreasoningJSON / Tool use
Kimi K2.5Current

Use when the workload needs code, 256k context, and JSON / Tool use.

2026-03code256k contextJSON / Tool use

Release Timeline

2 release groups
2026-07
1 current
Kimi K3
1.05m contextreasoningJSON / Tool use
Current
2026-03
1 current
Kimi K2.5
code256k contextJSON / Tool use
Current

Specifications(2 models)

Kimi model specifications comparison
ModelReleasedContextParametersVisionMultimodalReasoningJSON / Tool useStructured Outputs
Kimi K32026-071.05m2.8T total; 16 of 896 experts activeYesYesYesYesYes
Kimi K2.52026-03256k1T (MoE, 384 experts)YesYesNoYesYes

Pricing

Kimi model pricing by provider
ModelProviderInput / 1MOutput / 1MType
Kimi K2.5OpenRouter$0.44$2Serverless
Kimi K2.5Together AI$0.5$2.8Serverless
Kimi K2.5Fireworks AI$0.6$3Serverless
Kimi K2.5AWS Bedrock$0.6$3Serverless
Kimi K2.5Replicate API$0.6$3Serverless
Kimi K2.5Vercel AI Gateway$0.6$3Serverless
Kimi K2.5Novita AI$0.6$3Serverless
Kimi K3Moonshot AI Kimi$3$15Serverless

Popular comparisons in this family