Inference
EN
Moonshot AI

Kimi K2.6

Kimi K2.6 is a reasoning model from Moonshot AI. 46 benchmarks count toward its index, in 7 categories.

  • Open weights
  • Reasoning
  • Input: text · image
  • Released 2026-04
RankedEnough results for an index.
IndexAll categories, out of 100.58.2±2.9Median 51.0
CoverageIndex weight that has results.95%
SpeedOutput tokens per second.81/sMedian 67/s
First tokenSeconds to the first output token.1.0 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$0.65Median $0.60
Output / 1MUS dollars per 1M output tokens.$3.41Median $2.55
ContextMaximum tokens in one request.262KMedian 262K
EloLMArena rating and rank.1455±4 #46Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
moonshotai/kimi-k2.6
Max outputMaximum output tokens per request.
236K
Knowledge cutoffLast date of training data.
—
ToolsTool calls on OpenRouter.
Yes
JSON outputOutput follows a given JSON schema.
Yes
Cache read / 1MUS dollars per 1M cached tokens.
$0.15
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.
13 providers
ProviderTypeMaker, cloud, host or router.InputOutputContext
ChutesHost$0.39$1.80262K
SiliconFlowHost$0.77$3.40262K
Novita AIHost$0.80$3.40262K
CoreWeaveHost$0.65$3.41262K
OpenRouterRouter$0.65$3.41262K
CrusoeDeprecatedHost$0.70$3.50262K
Deep InfraHost$0.75$3.50262K
GMI CloudHost$0.855$3.6066K
Cloudflare Workers AIDeprecatedCloud$0.95$4.00262K
DigitalOceanCloud$0.95$4.00262K
Microsoft AzureCloud$0.95$4.00262K
Moonshot AIMaker$0.95$4.00262K
PrivatemodeDeprecatedHost$1.79$8.94256K

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

Run locally

1026.88B parameters
16-bit
2157.4 GB
8-bit
1146.6 GB
4-bit
654.7 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.16 counted59.573 / 218
CodingCode writing and repair.7 counted59.373 / 209
ReasoningLogic problems and puzzles.6 counted53.890 / 227
MultimodalTasks with images and text.4 counted57.647 / 84
KnowledgeFacts and expert knowledge.5 counted56.185 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.1 counted60.0—
MathMath problems.7 counted60.050 / 168

Results

46 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
V*Multimodal96.9%1 / 14
τ²-Bench Tool-Agent-User EvaluationAgentic95.9%13 / 149
Artificial Analysis IFBenchInstruction76.0%15 / 146
FrontierMath-2025-02-28-PrivateMath39.0% ±2.99 / 69
Artificial Analysis GPQA DiamondKnowledge91.1%27 / 199
LiveCodeBenchCoding86.8% ±1.020 / 142
Artificial Analysis Long Context ReasoningReasoning81.0%35 / 217
Software Engineering Benchmark VerifiedCoding80.2%19 / 114
LiveCodeBench v6Coding89.6%10 / 59
GPQA diamond · Epoch AIKnowledge90.8% ±1.737 / 217
Scientific Code BenchmarkCoding52.2%6 / 35
Graduate-Level Google-Proof Q&AKnowledge90.5%26 / 141
OTIS Mock AIME 2024-2025Math96.1% ±2.437 / 200
MMLU ProKnowledge87.6% ±0.325 / 135
Terminal-Bench 2.0 · Vals AIAgentic57.3% ±5.313 / 66
AIME 2026Math96.4%7 / 34
MMMU ProMultimodal86.3% ±0.821 / 92
GPQA Diamond · Vals AIKnowledge89.1% ±2.031 / 135
Artificial Analysis MMMU-ProMultimodal79.4%26 / 113
FrontierMath-Tier-4-2025-07-01-PrivateMath14.6% ±5.113 / 55
Terminal-Bench 2.0 · BenchLMAgentic66.7%10 / 41
DeepSearchQAAgentic92.5%6 / 24
Artificial Analysis Humanity's Last ExamKnowledge37.5%57 / 222
Harvard-MIT Mathematics Tournament February 2026Math92.7%7 / 27
SWE-Bench verified · Epoch AICoding76.7% ±1.99 / 32
Gert Labs Composite Game BenchmarkAgentic56.8%15 / 52
GPQA Diamond · BenchLMKnowledge90.5%26 / 84
WideResearchAgentic80.8%5 / 16
Massive Multi-discipline Multimodal Understanding ProMultimodal79.4%17 / 53
Critical Physics TasksReasoning8.0%75 / 214
Artificial Analysis SciCodeCoding51.5%45 / 114
BrowseCompAgentic83.2%26 / 63
Artificial Analysis Omniscience AccuracyKnowledge32.6%72 / 164
ToolathlonAgentic50.0%10 / 22
SWE-bench ProCoding58.6%38 / 83
APEX-Agents-AAAgentic28.5%14 / 30
SWE-benchCoding76.2% ±1.941 / 87
GDPval-AA normalizedAgentic26.9%67 / 133
FrontierMath-Tiers-1-3-v2-PrivateMath57.2% ±2.947 / 90
OSWorld-VerifiedAgentic73.1%20 / 37
MathVisionMultimodal87.4%13 / 24
Humanity's Last ExamKnowledge34.7%42 / 77
Code MigrationCoding27.8% ±4.143 / 77
Claw-EvalAgentic62.3%24 / 42
Vibe Code Bench v1.1Coding37.9% ±4.976 / 123
FrontierMath-Tier-4-v2-PrivateMath25.6% ±7.145 / 72
SimpleQA VerifiedKnowledge34.9% ±1.558 / 91
CharXiv ReasoningMultimodal80.4%33 / 49
Chess PuzzlesReasoning20.0% ±4.064 / 90
Terminal-Bench 2.1 · Vals AIAgentic53.6% ±2.154 / 75
Mystery Game PuzzlesReasoning12.0% ±3.361 / 80
Terminal-Bench HardAgentic43.9%18 / 23
MCP AtlasAgentic55.9%41 / 52
OSWorld 2.0Agentic4.6%23 / 25
Furniture AssemblyReasoning21.7% ±5.235 / 37
EBR-benchReasoning2.4%30 / 31
FrontierMath-2025-02-28-PublicMath90.0% ±10.02 / 64
CorpFin v2Finance66.7% ±0.916 / 131
LegalBenchLegal84.7% ±0.430 / 149
TaxEval v2Finance74.7% ±0.831 / 143
Finance Agent v1.1Finance57.1% ±2.912 / 51
MortgageTaxFinance65.8% ±0.939 / 97
CaseLaw v2Legal61.2% ±2.122 / 53
EMBFinance57.9% ±2.739 / 75
ProgramBenchCoding0.0%34 / 65
MedScribeHealthcare78.1% ±1.867 / 109
MedCodeHealthcare40.1% ±2.068 / 106
Public Benefits Bench v1Public benefits53.4% ±1.39 / 13
Finance Agent (v2)Finance44.9% ±0.756 / 79
Tax Agent BenchFinance12.4% ±2.463 / 82
Legal Research BenchLegal15.9% ±2.570 / 90
Public Benefits Bench v1.1Public benefits56.6% ±1.339 / 50
MMAnswerBenchMath86.0%—
MMMU-Pro with PythonMultimodal80.1%—
cursorBench31Coding47.6%—
ResearchClawBenchAgentic18.0%—

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseEpoch AICC BYVals AINo licence statedmodels.devMIT

History

5 releases
Index59.2 → 58.2
23 Oct 202612 Nov 2026
Output / 1M$3.41 → $3.41
23 Oct 202612 Nov 2026

Questions

How much does Kimi K2.6 cost?
$0.65 per 1M input tokens, $3.41 per 1M output tokens.
How good is Kimi K2.6?
Its index is 58.2 out of 100. That is place 83 of 283 ranked models.
What is Kimi K2.6 best at?
Math: 60.0 out of 100, place 50 of 168.
How fast is Kimi K2.6?
81 output tokens per second, 1.0 s to the first token.
What is the context window of Kimi K2.6?
262K tokens.
Is Kimi K2.6 open source?
Yes. Its weights are published.
When was Kimi K2.6 released?
April 2026.

Badge

Kimi K2.6 on Inference360: place 83 of 283, index 58.2
HTML
<a href="https://inference360.ai/models/kimi-2-6"><img src="https://inference360.ai/badges/kimi-2-6.svg" alt="Kimi K2.6 on Inference360: place 83 of 283, index 58.2" height="20"></a>
Markdown
[![Kimi K2.6 on Inference360: place 83 of 283, index 58.2](https://inference360.ai/badges/kimi-2-6.svg)](https://inference360.ai/models/kimi-2-6)

Compare with

Claude Opus 4.658.0Beam 1.0 501B-A23B NVFP458.5Gemini 3 Pro57.9Kimi K2.7 Code57.7Beam 1.058.8

Lower price

dots3-note Preview63.4 · FreeApodex 1.1 Mini55.7 · FreeGLM-5.3-Flash63.1 · $0.078/1M

More from Moonshot AI

All models
Kimi LatestUnrankedKimi K2.7 Code57.7Kimi K2.5 (Reasoning)53.0Kimi K2.552.2Kimi K366.9