$ cat wiki/models/kimi-k3-1-pro.md
Kimi K3.1 Pro
Compared with
- Beam 1.0
- LFM2.5-24B-A2B
- EmbeddingGemma 2
- Mistral Large 4
- Kolibri-1
- GPT-6.1 Sol
- MiMo-V2.6-Pro
- Grok 4.7
- Ternary Bonsai 2 27B
- Fugu Max
- DeepSeek V4.1-Flash
- K2 Horizon
- Hy4 preview
- GLM-5.3-Flash
- Granite 4.2
- Ling-3.0-tiny
- Laguna S 2.1
- Inkling
- LongCat-2.0
- Gemini 4 Argon
- Claude Sonnet 5.5
- Claude Opus 5.5
- GPT-6 Luna
- GPT-6 Sol
- Gemini 3.8 Live
- Fugu Ultra v2
- Kimi K2.8 Preview
- GPT-Image-2.5 Flare
- GPT-Image-2.5 Sunburst
- Astra
- Gemini 3.8 Flash
- Muse Spark 1.3
- Claude Fable 5.1
- GLM-5.3
- Qwen 3.8 27B
- DeepSeek V4-Pro-0813
- Gemini 3.7 Flash
- Muse Glimmer
- Grok 4.6
- Muse Spark 1.2
- Qwen 3.8 Max
- DeepSeek V4-Flash
- Claude Opus 5
- Gemini 3.5 Flash-Lite
- Gemini 3.6 Flash
- DeepSeek V4
- GPT-5.6 Sol
- Grok 4.5
- Claude Sonnet 5
- GLM-5.2
- Claude Fable 5
- MiniMax M3
- Claude Opus 4.8
- Gemini 3.5 Flash
- Grok Build
- Claude Opus 4.7
- Muse Spark
Spec
| Attribute | Value |
|---|---|
| Developer | Moonshot AI |
| Released | 2026-10-23; hosted access (source) |
| Announced | 2026-10-23 (release date) |
| Context window | 1,048,576 tokens (source) |
| Pricing | $2.00/MTok input · $10.00/MTok output · $0.20/MTok cached input; cache writes $2.00/MTok (5m), $4.00/MTok (1h); excludes taxes (source) |
| License | unknown; full weights promised by 2026-11-04 (source) |
| Availability | Kimi.ai, mobile apps, Kimi Work, Kimi Code, Kimi API; enterprise access off by default at launch (source) |
Release Date
Hosted release on 2026-10-23, corroborated by the dated gateway announcement. Moonshot describes K3.1 Pro as a post-trained Kimi K3 with substantially more post-training compute and joint RL across hundreds of domains. Its “open-weight” description is prospective: the announcement promises full weights by 2026-11-04, rather than establishing a downloadable release. (source) (release date)
Benchmarks
Moonshot reports Terminal-Bench 4.0 65.7%, against 58.2% for GPT-6 Astra and 66.4% for Claude Opus 5.5. Kimi uses the Claude-Code harness; competitor scores are the maximum reported in their blogs or the official leaderboard. Unless otherwise stated, Kimi uses effort max, temperature 1.0, top-p 0.95. These are vendor-assembled comparisons, not a fresh uniform evaluation. (source)
Its internal Kimi Code Bench 3.0 Long-horizon rises from K3's 55.95% to 61.2%. On internal xdev-bench-hard, removing cyber-related mid-training data and RL environments lowers pass@3 from 90.5% for an unfiltered checkpoint to 81.0% for the final model. Runs have no hard timeout or output-token limit; the final score does not establish low absolute cyber capability. (source)
Use Cases
Moonshot targets coding, tool-using workflows, 3D creation and financial research. API customers can request Zero Data Retention. Its Highspeed serving variant is claimed to run 2-3× faster; separate endpoint rates are $4.00/MTok input, $20.00/MTok output, $0.20/MTok cached input, and cache writes $4.00/MTok (5m) or $8.00/MTok (1h). No separate Highspeed benchmark is established by the captured text. (source) (source)
Compared To
Kimi K3 is the post-training predecessor; Claude Opus 5.5 and Astra appear in Moonshot's comparisons. The announced cyber-data intervention connects the release to Open-Weights Policy Fight; the internal evaluation supports a relative reduction, not a guarantee about downstream deployment. (source)
Sources
- Moonshot announcement — launch claims and evaluation footnotes.
- Kimi pricing capture — exact context, token rates and cache-write tiers, read 2026-10-24.
- Dated gateway announcement — release-date corroboration.