Identify the model and its published limits
The UncensuredAI catalog identifies Kimi 3.1P Uncensored with the API model ID kimi-k3.1-pro-uncensored. It lists a 1,048,576-token context window and a maximum output of 131,072 tokens. The catalog describes it as useful for coding and software development and less suited to cybersecurity work.
These are published product details, not a benchmark performed by this guide. The context limit describes capacity; it does not guarantee perfect recall across a large repository or manuscript. Start with the relevant files or passages and check whether the answer uses them accurately.
Compare the options on matching quantities
These UncensuredAI rates were checked on 11 November 2026. All figures are USD per million tokens, with separate input and output rates.
| Edition | Input | Output |
|---|---|---|
| Kimi 3.1P Uncensored | $0.42 | $2.10 |
| GLM 5.3 Flash Abliterated | $0.22 | $0.80 |
| GLM 5.3 Abliterated | $0.70 | $1.95 |
| DeepSeek V4 Pro Uncensored | $0.35 | $0.65 |
Two million uncached input tokens and one million output tokens cost $2.94 on Kimi, compared with $1.24 on GLM 5.3 Flash Abliterated. These inference subtotals exclude caching and any other applicable costs. A price difference is a reason to test, not evidence that the models are interchangeable.
Use three tasks that reveal useful differences
For coding, provide a small representative module and ask for a change with explicit acceptance criteria. Inspect correctness and how much repair the output needs. For writing, ask for a constrained scene and a focused revision. For document analysis, require claims linked to passages you can check.
Keep the input and success criteria consistent between candidates. Record the tokens used, the response time you observe and whether the result is acceptable. Repeat enough examples to avoid choosing a model solely because of one unusually strong or weak answer.
Plan around operational details, not only model quality
The uncensored catalog warns that capacity is shared, HTTP 429 responses can occur, and fixed RPM or TPM quotas are not guaranteed. Your application should limit parallel requests and retry with backoff. A successful single request does not establish capacity for a busy customer-facing product.
Check context and output limits individually for each alternative. For example, the cited DeepSeek V4 Pro entry does not publish those two limits. Do not transfer Kimi’s specifications to it or assume that every model supports the same parameters because the API interface is familiar.
Choose an edition for a specific job
Kimi belongs to the catalog’s Uncensored collection. Original models available on the same platform retain their producers’ safeguards, while Abliterated entries are labeled separately. Treat each edition as a distinct candidate and evaluate the behavior you need rather than promising that any label guarantees unrestricted or flawless responses.
Give Kimi a practical audition. Open its current catalog entry, copy the exact ID and connect it to your application. Keep it where its results justify its cost, and use another model where your own evaluation shows a better fit.
Put the idea to work.
Explore the current UncensuredAI catalog, create an API key or discuss the models you want to offer through your own service.
Sources and scope
Prices are USD per million tokens unless stated otherwise. Sources checked 11 Nov 2026. This is a first-party publishing project; see our method and disclosure.