Inference
EN
Moonshot AI

Kimi K3.1 Pro

Kimi K3.1 Pro is a reasoning model from Moonshot AI in the Kimi K3.1 family. 35 benchmarks count toward its index, in 5 categories.

  • Smartest openHighest index with open weights.
  • Open weights
  • Reasoning
  • Input: text
  • Released 2026-10
RankedEnough results for an index.
IndexAll categories, out of 100.75.7±4.7Median 51.0
CoverageIndex weight that has results.80%
SpeedOutput tokens per second.43/sMedian 67/s
First tokenSeconds to the first output token.49.6 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$2.00batch $0.70Median $0.60
Output / 1MUS dollars per 1M output tokens.$10.00batch $6.00 Batch price per 1M output tokens.Median $2.55
ContextMaximum tokens in one request.1.05MMedian 262K
EloLMArena rating and rank.—Median 1422

Run locally

2780.11B parameters
16-bit
5839.2 GB
8-bit
3102.6 GB
4-bit
1770.7 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.21 counted76.77 / 218
CodingCode writing and repair.3 counted71.916 / 209
ReasoningLogic problems and puzzles.3 counted77.711 / 227
MultimodalTasks with images and text.5 counted72.08 / 84
KnowledgeFacts and expert knowledge.3 counted71.016 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.None counted——
MathMath problems.None counted——

Results

35 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
Artificial Analysis Long Context ReasoningReasoning86.3%3 / 217
GDPval-AA normalizedAgentic67.3%3 / 133
Humanity's Last ExamKnowledgeAbove fitFar above this model's index.70.0%2 / 77
Humanity's Last Exam with toolsAgentic70.0%1 / 36
Graduate-Level Google-Proof Q&AKnowledge94.2%5 / 141
Massive Multi-discipline Multimodal Understanding ProMultimodal84.1%2 / 53
OSWorld 2.0Agentic79.6%1 / 25
Artificial Analysis Terminal-Bench v4.0Agentic67.2%1 / 22
BrowseCompAgentic92.1%3 / 63
Artificial Analysis Humanity's Last ExamKnowledge52.5%11 / 222
Artificial Analysis CyberGym-E2E-AAAgenticAbove fitFar above this model's index.94.7%1 / 19
Artificial Analysis Cyber IndexAgenticAbove fitFar above this model's index.64.1%1 / 19
CharXiv Reasoning without toolsMultimodal89.0%1 / 17
CharXiv ReasoningMultimodal93.4%3 / 49
GPQA Diamond · BenchLMKnowledge94.2%6 / 84
Artificial Analysis SciCodeCoding60.8%9 / 114
Artificial Analysis MMMU-ProMultimodal83.9%9 / 113
DeepSWEAgentic75.5%5 / 59
Artificial Analysis GDP.pdfAgentic31.4%2 / 22
Critical Physics TasksReasoning24.6%22 / 214
Artificial Analysis Omniscience AccuracyKnowledge48.9%23 / 164
Artificial Analysis DeepsecBench-AAAgentic36.7%3 / 21
Humanity's Last Exam without toolsKnowledge51.1%8 / 53
APEX-AgentsAgentic69.4%2 / 13
Artificial Analysis AutomationBenchAgentic73.3%4 / 23
Medical Long Context Reasoning (MLCR-AA)Reasoning38.3%4 / 22
OfficeQA ProMultimodal67.4%4 / 16
Agents' Last ExamAgentic43.2%6 / 21
ZeroBenchMultimodal39.0%3 / 10
MazeBench: visual spatial reasoning in a 3D open worldAgentic2.0%15 / 48
Artificial Analysis CWE-Bench-AAAgentic60.8%6 / 19
OSWorld 2.1 (all 108 tasks)Agentic79.6%4 / 11
Bug Hunt BenchCoding36.0%11 / 25
Multimodal Knowledge Agent BenchmarkAgentic43.9%9 / 19
ExploitGymAgentic20.7%13 / 23
FrontierCode 2.0Coding19.6%22 / 27
Terminal-Bench-Science 0.1Agentic38.1%9 / 11
Artificial Analysis AnalystAgentAgentic38.8%23 / 26
CWE-bench v2Agentic45.3%10 / 11
BabyVision with PythonMultimodal95.4%—
BenchCAD Vision2Code voxel IoU with toolsMultimodal92.7%—
Chartography with image and code toolsMultimodal90.2%—
MMMU-Pro with PythonMultimodal90.0%—
GDP.pdf task-macro rubric pass rate with Python, Gemini judgeMultimodal85.7%—
BioMysteryBench (overall, v11 release)Agentic80.6%—
Kimi Code Bench v2Coding77.1%—
BenchCAD Vision2Code voxel IoU without toolsMultimodal72.8%—
Financial Research (Internal)Agentic70.4%—
PerceptionBench (Internal)Multimodal68.3%—
OfficeQA Pro v2Multimodal64.4%—
ZeroBench_main with PythonMultimodal63.0%—
Kimi Code Bench 3.0 Long-horizon (Internal)Coding61.2%—
Frontier FinanceAgentic58.6%—
Financial Modeling (Internal)Agentic57.6%—
SWE-MarathonCoding51.7%—
HANDBOOK.mdAgentic44.6%—
Agents' Last Exam v1.1, CLI subsetAgentic44.2%—
Business Operation (Internal)Agentic37.8%—
PostTrainBench v1.2Coding36.5%—
GDP.pdf strict pass rate, page images plus text, Gemini judgeMultimodal34.6%—
Harvey's Legal Agent Benchmark, public 1,251-task set, all-passAgentic23.7%—
Kimi Hallucination Benchmark Hard with tools (Internal)Agentic8.3%—
Kimi Hallucination Benchmark without tools (Internal)Knowledge5.4%—

Sources

BenchLMCC BY-NC 4.0

History

5 releases
Index77.7 → 75.7
29 Oct 202612 Nov 2026
Output / 1M$10.00 → $10.00
29 Oct 202612 Nov 2026

Questions

How much does Kimi K3.1 Pro cost?
$2.00 per 1M input tokens, $10.00 per 1M output tokens.
How good is Kimi K3.1 Pro?
Its index is 75.7 out of 100. That is place 11 of 283 ranked models.
What is Kimi K3.1 Pro best at?
Agentic: 76.7 out of 100, place 7 of 218.
How fast is Kimi K3.1 Pro?
43 output tokens per second, 49.6 s to the first token.
What is the context window of Kimi K3.1 Pro?
1.05M tokens.
Is Kimi K3.1 Pro open source?
Yes. Its weights are published.
When was Kimi K3.1 Pro released?
October 2026.

Badge

Kimi K3.1 Pro on Inference360: place 11 of 283, index 75.7
HTML
<a href="https://inference360.ai/models/kimi-3-1-pro"><img src="https://inference360.ai/badges/kimi-3-1-pro.svg" alt="Kimi K3.1 Pro on Inference360: place 11 of 283, index 75.7" height="20"></a>
Markdown
[![Kimi K3.1 Pro on Inference360: place 11 of 283, index 75.7](https://inference360.ai/badges/kimi-3-1-pro.svg)](https://inference360.ai/models/kimi-3-1-pro)

Compare with

Claude Fable 5.176.2GPT-5.5 Pro74.8GPT-6.1 Terra77.1Claude Sonnet 5.577.2Muse Wonder Full77.3

Lower price

GPT-6.1 Terra77.1 · $5.00/1M

More from Moonshot AI

All models
Kimi K366.9Kimi K2.7 Code57.7Kimi LatestUnrankedKimi K2.658.2Kimi K2.5 (Reasoning)53.0