Singularity Logs

THE MODEL, IN DETAIL / Moonshot

Kimi K3 (high)

moonshotai/kimi-k3Open weightsReasoninghigh configuration

Kimi K3 is a 2.8T parameter open-weight multimodal reasoning model from Moonshot AI. It is suited for complex coding, knowledge work, and long-horizon agentic workflows, and is particularly strong at...

OpenRouter
Source-linked model intelligenceCatalog added 16 Jul 2026Specifications observed 28 Oct 2026

Specifications and providers describe MoonshotAI: Kimi K3. Evaluation scores retain the exact high configuration.

GPQA Diamond

91.9%

Source-reported

Context window

1M

943.7K max output tokens

Input price

$0.192

USD / million input tokens

Output price

$20

USD / million output tokens

Inference providers

21

30 provider routes observed

Measured output speed

—

No model-level measurement

01 / PERFORMANCE

A clearer view of capability.

Read the score, understand the context, and explore where this model sits among its peers.

Performance fingerprint

Individual published accuracy scores · not a composite index

At least three published accuracy scores are needed for this view.

Performance meets price

GPQA Diamond · same published source and version

Higher score · lower cost80$085$2.87590$5.7595$8.625100$11.5Gemini 4 Argon (high) GPQA Diamond: 96.4 Input: $2/M tokensGPT-6 Astra (max) GPQA Diamond: 95.8 Input: $10/M tokensClaude Sonnet 5.5 (max) GPQA Diamond: 95.6 Input: $2/M tokensKimi K3 (max) GPQA Diamond: 93.1 Input: $0.192/M tokensClaude Opus 5 GPQA Diamond: 92.9 Input: $5/M tokensGemini 3.5 Flash GPQA Diamond: 92.8 Input: $1.5/M tokensGrok 4.7 (xhigh) GPQA Diamond: 92.7 Input: $2/M tokensQwen3.8 Max (0902) (xhigh) GPQA Diamond: 92.3 Input: $2/M tokensKimi K3 (high) GPQA Diamond: 91.9 Input: $0.192/M tokensKimi K3 (high)GLM-5.2 GPQA Diamond: 91.9 Input: $0.0068/M tokensDeepSeek V4 Pro 0813 (max) GPQA Diamond: 91.7 Input: $0.66/M tokensGPT-5.6 Luna (max) GPQA Diamond: 91.6 Input: $0.2/M tokensGPT-5.2 (xhigh) GPQA Diamond: 91.4 Input: $1.75/M tokensClaude Opus 4.8 GPQA Diamond: 91 Input: $5/M tokensDeepSeek V4 Flash 0731 (max) GPQA Diamond: 91 Input: $0.119/M tokensKimi K2.6 GPQA Diamond: 90.8 Input: $0.65/M tokensInput price · USD / million tokens
Google DeepMindOpenAIAnthropicMoonshotxAIAlibabaZ.ai (Zhipu AI)DeepSeek

In good company

GPQA Diamond · Epoch AI · benchmark data

96.4
Google DeepMind
95.8
OpenAI
95.6
Anthropic
93.1
Moonshot
92.9
Anthropic
92.8
Google DeepMind
92.7
xAI
92.7
Alibaba
92.6
Google DeepMind
91.9
This model

Rank #26 in this published cohort. 325 models in the source cohort. Showing selected nearby models. Evaluation settings can differ.

Evidence, by task.

Coverage is not a capability score.

2published results
1 evaluation source
Reasoning2 evaluations
CodingAwaiting evidence
Mathematics1 evaluation
VisionAwaiting evidence
AgenticAwaiting evidence
Long contextAwaiting evidence

Documented features, task evaluations, and preference rankings describe different things. Missing coverage does not imply a model cannot perform a task.

Explore our methodology

02 / BENCHMARK EXPLORER

The numbers behind the model.

Original results, exact versions, and a direct path to the evidence behind every score.

2 published results
020406080100GPQA Diamond: 91.9% · Epoch AI · independent runs · Epoch AI · benchmark data91.9%GPQA DiamondHigher is betterOTIS Mock AIME 2024–2025: 93.3% · Epoch AI · independent runs · Epoch AI · benchmark data93.3%OTIS Mock AIME 2…Higher is better

2 percentage-based evaluations. Each benchmark measures a different task; compare results within the same evaluation and protocol.

Published benchmark results for Kimi K3 (high)
BenchmarkScoreVersion & metricSourcePublishedEvidence
Epoch AI · independent runsAccuracy · Higher is betterEpoch AI · benchmark dataSource-reportedNot available
Epoch AI · independent runsAccuracy · Higher is betterEpoch AI · benchmark dataSource-reportedNot available

03 / COST & VALUE

Know what your workload costs.

Explore token prices, see where the spend goes, and model a workload of your own.

Input$0.192per million tokens
Output$20per million tokens
Cache read$0.182per million tokens
Cache write—per million tokens

Workload cost calculator

Change the numbers. See the trade-offs.

Token estimate only. Excludes cache writes, tools, and other provider fees. Actual token usage and provider limits vary.

Your estimated spend

1,000 requests · 11M tokens

$21.92estimated total
Uncached input$1.92
Cached input$0
Output$20
Per request$0.0219
Without cache discount$21.92
Estimated cache savings$0

04 / INFERENCE NETWORK

One model. Different ways to run it.

Compare inference routes by price, context, quantization, and reported availability.

21 providers · 30 routes
Provider routes for Kimi K3 (high), observed 28 Oct 2026
Provider / routeInput / MOutput / MCache / MContextOutput limitSpeedLatencyAvailability · 1dRates
RRelaceFP4$0.151$9.7224$0.151M943.7K——99.93%
WWaferMXFP4$0.189$20$0.1791M943.7K——100%
SSail ResearchFP4$0.23$11$0.1151M943.7K——100%
SSail ResearchFP4$0.23$11$0.1151M943.7K——100%
IInferenceNetFP4$0.239$10.8$0.211M943.7K——100%
IInferenceNetFP4$0.239$10.8$0.211M943.7K——100%
WWaferMXFP4$0.36$11.9$0.3591M943.7K——100%
IInferenceNetFP4 · priority$0.5$11.88$0.351M943.7K——100%
IInferenceNetFP4 · priority$0.5$11.88$0.351M943.7K——100%
WWaferMXFP4 · priority$0.5$22$0.41M943.7K——100%
MMakoraFP4$0.65$15$0.231M943.7K——99.84%
MMorphFP8$1.035$14.626$0.2921M943.7K——100%
AAkashMLFP4$1.2$14$1.21M943.7K——98.4%
DDecartMXFP4$1.65$8.25$0.1651M943.7K——99.79%
CChutesMXFP4$1.85$9.5$0.1851M65.5K——100%
FFireworksQuantization not disclosed$2.1$10.5$0.211M943.7K——95.61%
PPhalaQuantization not disclosed$2.55$12.75$0.2551M943.7K——100%
DDigitalOceanQuantization not disclosed$2.55$12.95$0.2551M943.7K——100%
DDeepInfraMXFP4$2.85$14.25$0.2851M943.7K——100%
PParetoAlphaQuantization not disclosed$3$15$0.31M943.7K——98.15%
AAmazon BedrockQuantization not disclosed$3$15$0.31M128K——99.78%
PParasailFP4$3$15$0.31M943.7K——100%
MModalMXFP4$3$15$0.31M943.7K——100%
TTogetherQuantization not disclosed$3$15$0.31M943.7K——98.69%
FFireworksQuantization not disclosed$3$15$0.31M943.7K——100%
BBaseTenFP8$3$15$0.31M262.1K——100%
MMoonshot AIQuantization not disclosed$3$15$0.31M943.7K——100%
AAlibabaQuantization not disclosed$3.45$17.25$0.3451M131.1K——99.35%
FFireworksQuantization not disclosed$4.5$22.5$0.451M943.7K——72.7%
FFireworksQuantization not disclosed · priority$4.5$22.5$0.451M943.7K——96.33%

Snapshot observed 28 Oct 2026. Speed and latency are the reported 30-minute median when available. This is a dated snapshot, not live monitoring.

Provider price comparison

USD per million tokens · input and output rates

InputOutput
Relacefp4
Input
$0.151
Output
$9.7224
Wafermxfp4
Input
$0.189
Output
$20
Sail Researchfp4
Input
$0.23
Output
$11
Sail Researchfp4
Input
$0.23
Output
$11
InferenceNetfp4
Input
$0.239
Output
$10.8
InferenceNetfp4
Input
$0.239
Output
$10.8
Wafermxfp4
Input
$0.36
Output
$11.9
InferenceNetfp4 · priority
Input
$0.5
Output
$11.88
InferenceNetfp4 · priority
Input
$0.5
Output
$11.88
Wafermxfp4 · priority
Input
$0.5
Output
$22

Reported availability

Provider availability over the selected observation window

Relace · FP499.93%
Wafer · MXFP4100%
Sail Research · FP4100%
Sail Research · FP4100%
InferenceNet · FP4100%
InferenceNet · FP4100%
Wafer · MXFP4100%
InferenceNet · priority · FP4100%
InferenceNet · priority · FP4100%
Wafer · priority · MXFP4100%

Output speed

Reported 30-minute p50 · tokens per second

No provider throughput measurements were returned in this snapshot.

Response latency

Reported 30-minute p50 · seconds

No provider latency measurements were returned in this snapshot.

Context by provider

Actual route limits can differ from the catalog maximum

Relace · FP41M
Wafer · MXFP41M
Sail Research · FP41M
Sail Research · FP41M
InferenceNet · FP41M
InferenceNet · FP41M
Wafer · MXFP41M
InferenceNet · priority · FP41M
InferenceNet · priority · FP41M
Wafer · priority · MXFP41M

05 / CAPABILITIES & MODALITIES

What goes in. What comes out.

A map of documented interfaces and API features, separate from benchmark performance.

INPUT MODALITIES

TextImageVideo
Kimi K3 (high)Reasoning supported

OUTPUT MODALITIES

Text

Feature support

Listed catalog API parameters and documented pricing features

ReasoningSupported
Tool callingSupported
Tool choiceSupported
Structured outputsSupported
Response formatSupported
Reasoning detailsSupported
Seed controlSupported
Stop sequencesSupported
Prompt cachingSupported
TemperatureSupported

Room to work

Context capacity and completion limits

1Mtotal context tokens
Context window
943.7Kmax output

Output tokens share the total context budget. The maximum completion length is a separate limit, and individual provider routes may have smaller limits.

Catalog context
1,048,576 tokens
Maximum output
943,718 tokens
Top-provider context
1,048,576 tokens

Features vary by route

Check the provider before depending on a specific API feature.

API feature support by provider
Provider routeReasoningToolsStructured outputResponse formatImplicit caching
Relace · FP4SupportedSupportedNot listedNot listedSupported
Wafer · MXFP4SupportedSupportedSupportedSupportedSupported
Sail Research · FP4SupportedNot listedSupportedSupportedSupported
Sail Research · FP4SupportedSupportedSupportedSupportedSupported
InferenceNet · FP4SupportedSupportedSupportedSupportedSupported
InferenceNet · FP4SupportedSupportedSupportedSupportedSupported
Wafer · MXFP4SupportedSupportedSupportedSupportedSupported
InferenceNet · priority · FP4SupportedSupportedSupportedSupportedSupported
InferenceNet · priority · FP4SupportedSupportedSupportedSupportedSupported
Wafer · priority · MXFP4SupportedSupportedSupportedSupportedSupported
Makora · FP4SupportedSupportedSupportedSupportedSupported
Morph · FP8SupportedSupportedSupportedSupportedSupported
AkashML · FP4SupportedSupportedSupportedSupportedSupported
Decart · MXFP4SupportedSupportedSupportedSupportedSupported
Chutes · MXFP4SupportedNot listedSupportedSupportedSupported
FireworksSupportedSupportedSupportedSupportedSupported
PhalaSupportedSupportedSupportedSupportedSupported
DigitalOceanSupportedSupportedSupportedSupportedSupported
DeepInfra · MXFP4SupportedSupportedSupportedSupportedSupported
ParetoAlphaSupportedSupportedSupportedSupportedSupported
Amazon BedrockSupportedSupportedNot listedNot listedNot listed
Parasail · FP4SupportedSupportedSupportedSupportedSupported
Modal · MXFP4SupportedSupportedSupportedSupportedSupported
TogetherSupportedSupportedSupportedSupportedSupported
FireworksSupportedSupportedSupportedSupportedSupported
BaseTen · FP8SupportedSupportedNot listedNot listedSupported
Moonshot AISupportedSupportedSupportedSupportedSupported
AlibabaSupportedSupportedSupportedSupportedSupported
FireworksSupportedSupportedSupportedSupportedSupported
Fireworks · prioritySupportedSupportedSupportedSupportedSupported

06 / UNDER THE HOOD

The complete specification sheet.

Exact identities, model architecture, limits, and the defaults that shape a request.

Identity & architecture

Developer
Moonshot
Exact identity
kimi-k3_high
Subject
Model Configuration
Access
Open weights
License
Not disclosed
Parameters
Not disclosed
Tokenizer
Other
Instruction format
Not disclosed

Reasoning & controls

Reasoning
Supported
Mandatory reasoning
No
Reasoning by default
Enabled
Published configuration
high
Supported reasoning efforts
max, high, low, none
Default reasoning effort
max
API parameters
22
Provider moderation
No
Quantization variants
fp4, mxfp4, fp8

Dates & limits

Catalog added
16 Jul 2026
Knowledge cutoff
Not disclosed
Catalog expiration
Not documented
Specifications observed
28 Oct 2026
Providers observed
28 Oct 2026
Request limits
Provider-specific / not supplied
Canonical catalog ID
moonshotai/kimi-k3-20260715

API parameters & defaults

Availability is catalog-wide. Defaults can vary across providers.

frequency_penaltySupported · default not supplied
include_reasoningSupported · default not supplied
logit_biasSupported · default not supplied
logprobsSupported · default not supplied
max_completion_tokensSupported · default not supplied
max_tokensSupported · default not supplied
min_pSupported · default not supplied
parallel_tool_callsSupported · default not supplied
presence_penaltySupported · default not supplied
reasoningSupported · default not supplied
reasoning_effortSupported · default not supplied
repetition_penaltySupported · default not supplied
response_formatSupported · default not supplied
seedSupported · default not supplied
stopSupported · default not supplied
structured_outputsSupported · default not supplied
temperatureSupported · default not supplied
tool_choiceSupported · default not supplied
toolsSupported · default not supplied
top_kSupported · default not supplied
top_logprobsSupported · default not supplied
top_pDefault: 0.95

07 / BUILD WITH THIS MODEL

From exploration to your first request.

Use the documented catalog identifier and keep your API key in an environment variable.

Make an API request

const response = await fetch(
  "https://openrouter.ai/api/v1/chat/completions",
  {
    method: "POST",
    headers: {
      Authorization: `Bearer ${process.env.OPENROUTER_API_KEY}`,
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      "model": "moonshotai/kimi-k3",
      "messages": [
        {
          "role": "user",
          "content": "Explain the trade-offs in this design."
        }
      ],
      "reasoning": {
        "effort": "high"
      }
    }),
  },
);
if (!response.ok) throw new Error(`HTTP ${response.status}`);
const data = await response.json();
console.log(data.choices[0].message.content);

08 / PROVENANCE & HISTORY

Every number has a story.

Follow the source, check the observation date, and distinguish published results from model specifications.

The observation trail

Recorded events, in chronological order

Added to the model catalog

Catalog creation date, as recorded by OpenRouter.

Provider routes observed

30 inference routes in this snapshot.

Model specifications observed

Catalog pricing, limits, and supported parameters captured.

Observation dates describe when data was captured. They do not represent newly run evaluations or continuous performance monitoring.

Published observation ledger

Historical evidence, with source and version preserved.

Published observation history for Kimi K3 (high)
ObservedBenchmarkScoreVersionSourceVerification
28 Oct 2026OTIS Mock AIME 2024–2025Epoch AI · independent runsEpoch AI · benchmark dataSource-reported
28 Oct 2026GPQA DiamondEpoch AI · independent runsEpoch AI · benchmark dataSource-reported
28 Oct 2026OTIS Mock AIME 2024–2025Epoch AI · independent runsEpoch AI · benchmark dataSource-reported
28 Oct 2026GPQA DiamondEpoch AI · independent runsEpoch AI · benchmark dataSource-reported
28 Oct 2026OTIS Mock AIME 2024–2025Epoch AI · independent runsEpoch AI · benchmark dataSource-reported
28 Oct 2026GPQA DiamondEpoch AI · independent runsEpoch AI · benchmark dataSource-reported

09 / A FEW MORE DETAILS

Questions worth asking.

Quick answers, drawn from this model’s connected specifications.

How much does Kimi K3 (high) cost?

Catalog starting rates are $0.192 per million input tokens and $20 per million output tokens. Cache-read pricing is $0.182 per million tokens. Compare provider routes and service tiers above for their specific prices.

How much context can it handle?

1,048,576 tokens in the catalog context window. The catalog lists a maximum of 943,718 completion tokens. Provider limits can differ.

Does it support tools and structured output?

Tool calling: listed as supported. Structured outputs: listed as supported. See the provider feature matrix for route-specific differences.

How should I interpret these benchmark results?

Read the benchmark name, exact version, metric, source, and model configuration together. Different evaluation settings can change the result. Preference ratings and source-defined composite indices remain separate from task accuracy.

Are these prices and provider measurements live?

This page uses dated source snapshots. Catalog data was observed on 28 Oct 2026. Provider routes were observed on 28 Oct 2026. Missing measurements are not estimated or treated as zero.

KEEP EXPLORING

More from Moonshot.

Explore related identities from the same developer, or take them into a side-by-side comparison.

THE RIGHT MODEL STARTS WITH THE RIGHT CONTEXT.

Put the details side by side.

Explore the trade-offs in a direct model comparison.

Compare Kimi K3 (high)
Source-backed evidence. Different tasks. A clearer picture.
AboutMethodologyBenchmark dashboard