MODEL DETAIL · Moonshot AI

Kimi K3

Source-linked record by Moonshot AI

#3 OVERALL

Kimi K3 supports text, image, video, tools, reasoning workflows. Published fields are kept separate from benchmark evidence and community signals.

Fit score74.1Contextual · fit-v1

Coding76.2Database signal

Reasoning93.5Database signal

Context window1048576Published limit

Input price$0.73Per 1M tokens

Released2026-07-16Source date

DATABASE MODEL RECORD

What this record says.

Kimi K3 is cataloged by Moonshot AI with source-linked capabilities, pricing, release metadata, and explicitly separated evaluation evidence.

Source snapshot fetchedConfidence: N/A

TextImageVideoToolsReasoningOpen weights

EVALUATION EVIDENCE

Only model-specific published scores are shown here. Missing values remain N/A and are not estimated from a composite score.

Benchmark	Score	Status
AutoBenchPublished Best for Work Automations (AutoBench) scores.	N/A	N/A	View source ↗
BrowseCompPublished Best in Browsing (BrowseComp) scores.	91.2	Published score	View source ↗
SWE-Bench VerifiedPublished Best in Agentic Coding (SWE Bench) scores.	N/A	N/A	View source ↗
OSWorld-VerifiedPublished Best in Computer Use (OSWorld) scores.	N/A	N/A	View source ↗
Humanity's Last ExamPublished Best Overall (Humanity's Last Exam) scores.	56.0	Published score	View source ↗

This record’s overall fit, coding, and reasoning fields are separate catalog signals, not substitutes for benchmark results.

PUBLISHED PRICING

Input / 1M tokens$0.73

Output / 1M tokens$10.00

Context window1048576

Knowledge cutoffN/A

LicenseOpen weights

Pricing is shown as published by the catalog record. Confirm provider terms before production use.

FIT CONTEXT

N/A

Capabilities: Text · Image · Video · Tools · Reasoning. This is descriptive metadata, not a guarantee of task performance.

RANK HISTORY

No historical rank snapshots are recorded for this model.

RELATED RECORDS

EVIDENCE LAYERS

Published facts25 canonical fieldsCapabilities, limits, pricing, license, and release metadata.

Published benchmarks8 reported scoresExternal evaluations with their own source and version.

TopLLM evaluationNot availableTopLLM first-party task evaluations will be published separately from external benchmark results.