Kimi K3.1 Pro open weights released: what shipped, what the licence allows, and where it ranks.
Kimi K3.1 Pro open weights are now public: Moonshot AI released the weights on November 4, 2026, on Hugging Face, under the Kimi K3.1 Pro License, with the API price unchanged. This post sets out exactly what was released, translates the licence into plain terms, recomputes the open-weight field with this model included, and separates what changes for API users from what changes for self-hosters.
By AI Pro League · Published
Kimi K3.1 Pro open weights landed on November 4, 2026, and our AI lab release tracker changelog records the change today: the model’s open-weights flag in our database moved from false to true and its licence is now recorded as the Kimi K3.1 Pro License. Kimi K3.1 Pro itself is not new — it entered our model database at its API launch on Oct 23, 2026 — what is new is that anyone can now download and run the weights themselves. The release graduates the “Kimi K3.1 Pro open weights” item from our Upcoming tracker, which had the weights due by November 4 since the model’s launch-day record.
What was released: Kimi K3.1 Pro open weights on Hugging Face
The weights are published as moonshotai/Kimi-K3.1-Pro on Hugging Face — Moonshot’s official model card, and the authoritative source our changelog entry cites. Per that card, as recorded in our changelog: the download is roughly 1.56 TB across 96 files, with no access gate. The model behind the weights is the one already on its model page — Kimi K3.1 Pro — with a 1M-token context window, a maximum output of 131,072 tokens, and text, image and video inputs with text output, as recorded in our database.
The Kimi K3.1 Pro licence, in plain terms
The weights ship under the Kimi K3.1 Pro License, described on the model card as MIT-style — a permissive base, with two scale conditions that matter and that our changelog records verbatim from the card:
- Model-as-a-service at scale needs a separate agreement. A model-as-a-service business above $20M in group revenue over 12 months needs a separate agreement with Moonshot.
- Very large products must name the model. Products above 100M monthly users or $20M in monthly revenue must prominently display the model name.
Below those thresholds, the plain-terms reading is simple: download, run and build — but read the licence text on the model card before a commercial launch, because “open-weight” is not one licence. The 30 open-weights models we track ship under 13 different licences, and our best open-weight AI models ranking treats the licence column as part of the ranking, not fine print.
Where Kimi K3.1 Pro ranks among open-weight models — our recomputed cut
With this release, the open-weight field we track grows to 30 models, of which 21 carry a League Score. Recomputing that field with Kimi K3.1 Pro included: it enters at #5 among scored open-weight models, with a League Score of 69.4 (#18 overall). That score rests on a single recorded component — LiveBench overall at 71.3 measured Oct 26, 2026 — so its score coverage is 40% of the five-benchmark set, the caveat our score coverage explainer sets out. On price, its blended rate of $4.00 per million tokens sits above the scored open field’s median blended rate of $1.98 — this is a frontier-class open release, not a budget one. The full field lives in the Open-Weight Division standings and our best open-source AI models picks.
| # | Model | Maker | League Score | Coverage | Blended $/1M | Licence |
|---|---|---|---|---|---|---|
| 1 | DeepSeek V4.1 Flash | DeepSeek | 82.0 | 55% | $0.52 | MIT |
| 2 | DeepSeek V4.1 Flash 1017 | DeepSeek | 72.8 | 40% | $0.52 | MIT |
| 3 | Kimi K3 | Moonshot AI | 70.6 | 70% | $6.00 | Kimi K3 License |
| 4 | Kimi K2.6 | Moonshot AI | 69.7 | 35% | $1.71 | Modified MIT |
| 5 | Kimi K3.1 Pro | Moonshot AI | 69.4 | 40% | $4.00 | Kimi K3.1 Pro License |
| 6 | Qwen3.8-Max | Alibaba Qwen | 65.0 | 65% | $3.00 | Qwen3.8-Max License |
| 7 | DeepSeek V4 Pro (0813) | DeepSeek | 63.2 | 60% | $1.98 | MIT |
| 8 | Qwen3.5-397B-A17B | Alibaba Qwen | 60.7 | 35% | — | Apache-2.0 |
| 9 | Mistral Medium 3.5 | Mistral AI | 59.1 | 20% | $3.00 | Modified MIT |
| 10 | Llama 4 Maverick | Meta | 47.7 | 15% | — | Llama 4 Community License |
What changes for API users vs self-hosters
- API users: nothing changes today. The API price is unchanged at $2.00 input / $10.00 output per million tokens, with cached input at $0.20 — the rates on our AI model pricing index, checked Oct 24, 2026 against Moonshot’s pricing documentation. On a worked workload of 10M input + 2M output tokens the hosted bill is $40.00, against $5.40 for the open leader, DeepSeek V4.1 Flash.
- Self-hosters: everything changes. The model is now downloadable and runnable on your own infrastructure under the licence above — the bill becomes hardware instead of tokens, and the roughly 1.56 TB download is the honest scale warning: this is a datacentre-class artefact, not a laptop model.
- Choosing between the two Kimi flagships? Our head-to-head pages price and score the neighbours: Kimi K3 vs Kimi K3.1 Pro compared and Kimi K2.6 vs Kimi K3.1 Pro compared.
What the open-weights release does NOT change
- The League Score. It stays at 69.4 — the release adds a download, not a benchmark result; the recorded component behind the score is unchanged.
- The API price. Unchanged, as above — this is a weights event, not a pricing event.
- The hosted roadmap around it. The K2 series is still retiring on its published timetable — see the Kimi K2 retirement analysis — and those models’ own open weights remain the self-hosting route that outlives that deadline, the same pattern now extended to the current flagship.
The wider pattern is in our AI model rankings: labs increasingly ship the API first and the weights on a dated promise — and, as here and with Mistral Large 4 last month, keep it. A ranking that tracks open-weights status alongside scores is how a release like this shows up the day it lands.