Kimi K3
Released
2026-07-14
Last refreshed
2026-10-08
Status
Researched 52d ago
Open weightsCommercial use: conditionalMultimodalRAGAgentsLong contextVisionJSON / Tool useHighlight
Moonshot AI releases · 6 in the last 12 months · this family litChangelog →
Specifications
- Family
- Kimi
- Released
- 2026-07-14
- Context
- 1M
- Max output
- 1M
- Parameters
- 2.8T total; 16 of 896 experts active
- Architecture
- Mixture of Experts
- Specialization
- general
- Openness
- Open weights
- License
- Kimi K3 LicenseCommercial use: conditional
- Weights
- Available
- Code
- Unknown
Created by
Pricing
Output / 1M
$15.00
Input / 1M
$3.00
Cheapest of 3 routes · Featherless
Links
About
Kimi K3 is Moonshot AI's 2.8-trillion-parameter flagship multimodal model for long-horizon coding, knowledge work, deep reasoning, and agentic workflows. It uses Kimi Delta Attention, Attention Residuals, and a sparse MoE design (16 of 896 experts active), supports a 1,048,576-token context window, text/image/video input, always-on reasoning, ToolCalls, strict JSON Schema structured output, automatic context caching, and partial mode through Moonshot's OpenAI-compatible API.
Providers(3)
Compare all 3Compare API pricing across 3 tracked provider routes for input and output tokens, batch, and cached reads when available.
| Provider | Input / 1M | Output / 1M | Cache | Route |
|---|---|---|---|---|
| Featherless | $3.00 | $15.00 | - | Serverless |
| Moonshot AI Kimi | $3.00 | $15.00 | read $0.300 | Serverless |
| OpenRouter | $3.00 | $15.00 | read $0.300 | Serverless |
Capabilities
VisionMultimodalReasoningJSON / Tool useStructured OutputsPrompt Caching
Benchmark peer barsfor RAG
No task-mapped benchmark peers are available for this model yet.
Benchmark scores(17)
Scores are benchmark-specific and are direction-aware: the same numeric gap can mean very different outcomes across suites. Use the leaderboard context and this model's provider route to decide whether the winning margin is meaningful for your workload.
| Benchmark | Score | Version | Evaluation | Source |
|---|---|---|---|---|
| Program Bench | 77.8 | KimiCode harness; max reasoning effortObserved 2026-07-16 | — | Source |
| Terminal-Bench 2.1 | 88.3 | Terminal-Bench 2.1; KimiCode harness; max reasoning effortObserved 2026-07-16 | — | Source |
| Google-Proof Q&A | 93.5 | GPQA-Diamond; max reasoning effortObserved 2026-07-16 | — | Source |
| Humanity's Last Exam | 43.5 | HLE-Full without tools; max reasoning effortObserved 2026-07-16 | — | Source |
| Humanity's Last Exam — With Tools | 56.0 | HLE-Full with tools; max reasoning effortObserved 2026-07-16 | — | Source |
| BrowseComp | 91.2 | BrowseComp; 300K-token context compaction; max reasoning effortObserved 2026-07-16 | — | Source |
| DeepSearchQA | 95.0 | DeepSearchQA F1; max reasoning effortObserved 2026-07-16 | — | Source |
| Toolathlon | 73.2 | Toolathlon-Verified; max reasoning effortObserved 2026-07-16 | — | Source |
| MCP-Atlas | 84.2 | 500-task public subset; 100-turn limit; Gemini 3.1 Pro judge; max reasoning effortObserved 2026-07-16 | — | Source |
| AutomationBench | 30.8 | 600-task public subset; official GitHub setup; max reasoning effortObserved 2026-07-16 | — | Source |
| WorldVQA | 51.0 | WorldVQA ForceAnswer; max reasoning effort; five-run meanObserved 2026-07-16 | — | Source |
| MMMU Pro | 81.6 | MMMU-Pro; official protocol; three-run mean; max reasoning effortObserved 2026-07-16 | — | Source |
| CharXiv | 84.8 | CharXiv RQ; three-run mean; max reasoning effortObserved 2026-07-16 | — | Source |
| ARC-AGI-2 | 60.4 | ARC-AGI-2 semi-privateObserved 2026-10-01 | Configuration: kimi-k3 (Max) Harness: ARC Prize semi-private evaluation Evaluator: ARC Prize Foundation Confidence: confirmed Notes: ARC Prize official leaderboard data (ARC-AGI-2 semi-private), verified in https://arcprize.org/media/data/leaderboard/v2.json (generatedAt 2026-10-01T18:51Z) for modelDisplayName 'Kimi K3 (Max)'. Highest-scoring published effort selected. Discovered via Benchmark Heaven / Epoch AI mirror (lead only). | Source |
| DeepSWE 1.1 | 68.5 | DeepSWE 1.1Observed 2026-09-22 | Configuration: kimi-k3 (Max) Harness: mini-swe-agent (Datacurve DeepSWE 1.1) Evaluator: Datacurve Confidence: confirmed Notes: Datacurve DeepSWE 1.1 leaderboard (113 tasks, mini-swe-agent, Pass@1 over 4 runs). Verified on https://deepswe.datacurve.ai/ (page generated_at 2026-09-22; page shows integer %%). Precise value from Epoch AI mirror deepswe_external.csv (Source column = deepswe.datacurve.ai). Highest-scoring published effort selected. | Source |
| FrontierCode 1.1 Main | 44.2 | FrontierCode 1.1 MainObserved 2026-10-05 | Configuration: kimi-k3 (None) Harness: mini-swe-agent Evaluator: Cognition Confidence: confirmed Notes: Cognition FrontierCode 1.1 Main (100 tasks) mean@5 composite ('new_score'), verified in https://cognition.com/data/frontiercode-leaderboard/data.json (the data file behind https://cognition.com/frontiercode), retrieved 2026-10-05. Agent harness: mini-swe-agent. Highest-scoring published effort selected; timestamp is retrieval date. | Source |
| FrontierMath Tier 4 | 39.0 | FrontierMath Tier 4 v2 (private)Observed 2026-07-17 | Configuration: kimi-k3 (Max) Harness: Epoch AI FrontierMath Tier 4 v2 private set Evaluator: Epoch AI Confidence: confirmed Notes: Epoch AI-run evaluation (evaluator of record), from Epoch AI Benchmarking Hub data download https://epoch.ai/data/benchmark_data.zip (file frontiermath_tier_4_v2.csv, row 'kimi-k3_max', run started 2026-07-17). Epoch data is CC BY 4.0: credit 'Epoch AI, Capabilities & benchmarking, https://epoch.ai/benchmarks'. Highest-scoring Epoch configuration for this model selected; other configurations exist in the same file. Discovered via Benchmark Heaven (lead only). | Source |
Migration checks
No linked migration route is available for this model yet.
Created by
Pricing
Output / 1M
$15.00
Input / 1M
$3.00
Cheapest of 3 routes · Featherless
Links