Model record
Kimi K3 model overview
Kimi K3 is a api + app model from Moonshot AI. It was released on July 16, 2026. Its strongest case is one-million-token context, while buyers should account for full weights, the technical report, and announced vllm work remained future as of the checked date.
Kimi K3 benchmark snapshot
| Evaluation | Reported score | What it probes |
|---|---|---|
| MMLU-Pro | Not reported | Broad knowledge and multi-step reasoning |
| GPQA Diamond | Not reported | Graduate-level science reasoning |
| SWE-bench Verified | Not reported | Verified real-repository issue resolution |
| LiveCodeBench | Not reported | Contamination-aware competitive programming |
| SWE-Bench Pro | Not reported | Longer, harder professional software tasks |
| Artificial Analysis Intelligence Index | 57 | Composite third-party capability index; version matters |
Scores are percentages reported by model creators or benchmark maintainers under varying settings. A blank is preferable to an inferred result. See our methodology.
Kimi K3 strengths
- One-million-token context
- Native vision and long-horizon coding
- First-party app, coding-agent, and API access
Kimi K3 limitations
- Full weights, the technical report, and announced vLLM work remained future as of the checked date
- Requires preserved thinking history and may act too proactively without explicit constraints
- Provider reports a remaining UX gap versus its cited proprietary frontier peers
Compare Kimi K3
- Kimi K3 vs GPT-5.6 Sol — review fit, trade-offs, listed prices, and an evaluation plan.
Source record
Specifications and scores are linked to the best source located during review. Provider-reported results are not presented as third-party lab reproductions.
Source limitation: First-party Kimi source. Kimi.com, Kimi Work, Kimi Code, and API model kimi-k3 were available when checked. Official API prices are $0.30 per MTok for cache-hit input, $3.00 for cache-miss input, and $15.00 for output; inputPrice records cache-miss input because the schema has no cached-input field. Full weights, a technical report, and a KDA/vLLM contribution were announced for release by 2026-07-27 and were not treated as available on 2026-07-21. Active parameters and a weight license remain unavailable. Intelligence Index 57 is the Artificial Analysis v4.1 result: https://artificialanalysis.ai/articles/kimi-k3-achieves-3-in-the-artificial-analysis-intelligence-index-comparable-to-opus-4-8-and-gpt-5-5.
Read Kimi K3 technical blog ↗