Inference
EN
Moonshot AI

Kimi K2

Kimi K2 is a standard model from Moonshot AI. 17 benchmarks count toward its index, in 6 categories.

  • Open weights
  • Standard
  • Input: text
  • Released 2025-07
RankedEnough results for an index.
IndexAll categories, out of 100.39.6±5.4Median 51.0
CoverageIndex weight that has results.85%
SpeedOutput tokens per second.37/sMedian 67/s
First tokenSeconds to the first output token.1.3 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$0.57Median $0.60
Output / 1MUS dollars per 1M output tokens.$2.30Median $2.55
ContextMaximum tokens in one request.131KMedian 262K
EloLMArena rating and rank.—Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
moonshotai/kimi-k2
Max outputMaximum output tokens per request.
98K
Knowledge cutoffLast date of training data.
31 Dec 2024
ToolsTool calls on OpenRouter.
Yes
JSON outputOutput follows a given JSON schema.
No
Cache read / 1MUS dollars per 1M cached tokens.
—
WeightsLink to the published weights.
Hugging Face (opens in a new tab) ↗
RetiresDate OpenRouter removes the model.
—

Where to run

APIs that serve this model.
1 provider
ProviderTypeMaker, cloud, host or router.InputOutputContext
OpenRouterRouter$0.57$2.30131K

US dollars per 1M tokens, list price. From models.dev (opens in a new tab) (MIT).

Run locally

1026.41B parameters
16-bit
2156.5 GB
8-bit
1146.1 GB
4-bit
654.4 GB
Weights
Hugging Face (opens in a new tab) ↗

Memory need with a short context. The 4-bit copy needs more than 128 GB. Compare local models

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.1 counted44.9—
CodingCode writing and repair.2 counted39.0181 / 209
ReasoningLogic problems and puzzles.2 counted40.3161 / 227
MultimodalTasks with images and text.None counted——
KnowledgeFacts and expert knowledge.7 counted41.5183 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.2 counted35.947 / 63
MathMath problems.3 counted36.0134 / 168

Results

17 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
SWE-bench Verified · swebench.comCoding65.4%6 / 46
MATH-500 Problem SetMath97.4%7 / 27
Massive Multitask Language UnderstandingKnowledge89.5%9 / 34
Measuring Short-Form Factuality in Large Language ModelsKnowledge31.0%4 / 14
SWE-bench Verified · SWE-bench/experimentsCoding65.4%21 / 65
Instruction-Following EvalInstruction89.8%27 / 70
Massive Multitask Language Understanding ProfessionalKnowledge81.1%37 / 75
Artificial Analysis Omniscience AccuracyKnowledge27.4%90 / 164
Artificial Analysis GPQA DiamondKnowledge76.6%110 / 199
τ²-Bench Tool-Agent-User EvaluationAgentic61.1%85 / 149
Graduate-Level Google-Proof Q&AKnowledge75.1%81 / 141
Artificial Analysis IFBenchInstruction41.5%98 / 146
Artificial Analysis Long Context ReasoningReasoning53.0%148 / 217
Critical Physics TasksReasoning0.0%150 / 214
Software Engineering Benchmark VerifiedCoding65.8%80 / 114
American Invitational Mathematics Examination 2024Math69.6%19 / 27
Artificial Analysis Humanity's Last ExamKnowledge7.4%158 / 222
τ²-bench AirlineAgentic56.5%16 / 22
Artificial Analysis MATH-500Math97.1%7 / 9
τ²-bench TelecomAgentic65.8%14 / 18
American Invitational Mathematics Examination 2025Math49.5%44 / 53
LiveCodeBench v6Coding53.7%50 / 59
SuperGPQA: Scaling LLM Evaluation Across 285 Graduate DisciplinesKnowledge57.2%20 / 22
τ²-bench RetailAgentic70.6%21 / 23
Humanity's Last ExamKnowledge4.7%77 / 77

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseSierra ResearchMITSWE-bench teamNo licence statedmodels.devMIT

History

5 releases
Index40.5 → 39.6
23 Oct 202612 Nov 2026
Output / 1M$2.30 → $2.30
23 Oct 202612 Nov 2026

Questions

How much does Kimi K2 cost?
$0.57 per 1M input tokens, $2.30 per 1M output tokens.
How good is Kimi K2?
Its index is 39.6 out of 100. That is place 209 of 283 ranked models.
What is Kimi K2 best at?
Knowledge: 41.5 out of 100, place 183 of 268.
How fast is Kimi K2?
37 output tokens per second, 1.3 s to the first token.
What is the context window of Kimi K2?
131K tokens.
Is Kimi K2 open source?
Yes. Its weights are published.
When was Kimi K2 released?
July 2025.

Badge

Kimi K2 on Inference360: place 209 of 283, index 39.6
HTML
<a href="https://inference360.ai/models/kimi-k2"><img src="https://inference360.ai/badges/kimi-k2.svg" alt="Kimi K2 on Inference360: place 209 of 283, index 39.6" height="20"></a>
Markdown
[![Kimi K2 on Inference360: place 209 of 283, index 39.6](https://inference360.ai/badges/kimi-k2.svg)](https://inference360.ai/models/kimi-k2)

Compare with

Claude 4 Sonnet39.5LFM2.5-24B-A2B39.8LongCat-Flash-Lite-Sparse39.3o3-mini40.0dots3-note Preview63.4

Lower price

dots3-note Preview63.4 · FreeApodex 1.1 Mini55.7 · FreeNemotron 3 Ultra50.8 · Free

More from Moonshot AI

All models
Kimi K2 0905UnrankedKimi-Audio 7BUnrankedKimi K2 Thinking48.2Kimi K2.5 (Reasoning)53.0Kimi K2.552.2