---
title: "Kimi K3 jailbreak, or the uncensored Kimi K3.1 API"
canonical: https://pentestllm.ai/blog/kimi-k3-jailbreak
generated: 2026-11-18T03:47:04.526Z
source: https://pentestllm.ai/blog/kimi-k3-jailbreak
base_url: https://api.pentestllm.ai/v1
---

# Kimi K3 jailbreak: what the prompt costs, and the uncensored Kimi builds you call plain

Each message of a Kimi K3 jailbreak session pays for the jailbreak again. The uncensored Kimi builds, K3.1 Pro first, then K3 and K2.6, need none and share one key: live prices, context windows and a request to paste.

Published November 6, 2026 · prices as of November 17, 2026

A Kimi K3 jailbreak is text that tries to walk Moonshot AI’s Kimi past a refusal, and it is resent and charged with every message. The other way is a Kimi build served uncensored and called plain. PentestLLM carries 3 of them on one key, Kimi K3.1 Pro Uncensored first. Below: the live prices, a request to paste into a terminal, and what the catalog does and does not state about the build.

## What people mean by a Kimi K3 jailbreak

A jailbreak is a block of text. It tries to get a model that was trained to decline to answer anyway, through a persona, a cover story or a long preamble. Because a chat API remembers nothing between requests, that block goes out with every message of the session and is charged as input every time.

You will find no jailbreak prompts here. What this page gives you is the other path: the Kimi builds the catalog tags as uncensored, which you call plain, with a system prompt of your own or none at all.

What that block costs at today’s rate: 1,200 tokens on 100 messages is 120,000 input tokens, or $0.14 at the input price of Kimi K3.1 Pro Uncensored, spent on text that adds nothing after the first message.

## Three uncensored Kimi builds, one key

Pulled from the live catalog on 2026-11-17. Each name opens the build’s page, which has the settings for OpenCode, Claude Code and the other agents. The session column prices 100 replies of 300 tokens on 4,000 tokens of context, no jailbreak attached.

| Build | Input $/M | Output $/M | Context | Max output | One session | Daily check |
| --- | --- | --- | --- | --- | --- | --- |
| [Kimi K3.1 Pro Uncensored](https://pentestllm.ai/models/kimi-k3.1-pro-uncensored) | $1.164 | $5.82 | 1,048,576 | 131,072 | $0.64 | Failing requests this week (down) |
| [Kimi K3 Uncensored](https://pentestllm.ai/models/kimi-k3-uncensored) | $5.745 | $28.68 | 262,144 | 65,536 | $3.16 | Steady since Nov 3 |
| [Kimi K2.6 Uncensored](https://pentestllm.ai/models/kimi-k2.6-uncensored) | $1.17 | $7.34 | 262,144 | 65,536 | $0.69 | Not checked yet |

Worked at render: messages × context tokens × input rate, plus messages × 300 × output rate. The last column is the [daily real-model check](https://pentestllm.ai/proof): one fixed prompt to every model, each day, with a note of whether the reply still matches. A row marked failing means live requests to it are being rejected this week, whatever the fixed prompt returned.

Start with [Kimi K3.1 Pro Uncensored](https://pentestllm.ai/models/kimi-k3.1-pro-uncensored): $1.164 per million input tokens, $5.82 per million output tokens, cached input at $0.1164, a context window of 1,048,576 tokens and replies of up to 131,072 tokens. Kimi K3 Uncensored and Kimi K2.6 Uncensored sit on the same key under their own ids; an agent switches between them by editing one string.

## Send a request to Kimi K3.1 Pro Uncensored

The API takes the OpenAI chat format. Make a key in the [console](https://app.pentestllm.ai/console), point the client at `https://api.pentestllm.ai/v1` and put `kimi-k3.1-pro-uncensored` in the model field. Credit is prepaid from $10 and charged per token.

```
curl https://api.pentestllm.ai/v1/chat/completions \
  -H "Authorization: Bearer sk-ptl-live-…" \
  -H "Content-Type: application/json" \
  -d '{"model": "kimi-k3.1-pro-uncensored",
       "messages": [
         {"role": "system", "content": "You assist a penetration tester on an authorized engagement. Answer directly, no preamble."},
         {"role": "user", "content": "Review this login handler for injection and auth bypass issues and point at the lines that matter."}
       ],
       "max_tokens": 600}'
```

```python
from openai import OpenAI

client = OpenAI(base_url="https://api.pentestllm.ai/v1", api_key="sk-ptl-live-…")

completion = client.chat.completions.create(
    model="kimi-k3.1-pro-uncensored",
    messages=[{"role": "user", "content": "Hello"}],
)
print(completion.choices[0].message.content)
```

The system prompt names the engagement and the job. It does not negotiate with the model. OpenCode and Claude Code take the same base URL, key and id; the [OpenCode page](https://pentestllm.ai/agents/opencode) and the [Claude Code page](https://pentestllm.ai/agents/claude-code) show where each goes, the [docs page](https://pentestllm.ai/docs) covers the rest, and TypeScript and AI SDK examples are in the [SDK guide](https://pentestllm.ai/guides/sdks). With nothing installed, the home page gives a few demo turns and the [playground](https://pentestllm.ai/playground) runs the build on a key.

## What the catalog says about this build, and what it leaves out

- **Base model developer:** Moonshot AI, as recorded on the catalog row.
- **Uncensored:** `kimi-k3.1-pro-uncensored` carries the uncensored tag, which on this site means a build that answers without the stock model’s safety refusals. It does not carry the abliterated tag, and no weights or build notes have been published for it, so how it was made is not stated here. For builds that do carry the abliterated tag, see the [abliterated page](https://pentestllm.ai/abliterated).
- **Context:** 1,048,576 tokens, replies of up to 131,072 tokens. The whole context is resent with each message, so set an agent’s context limit to what the task needs rather than to the maximum.
- **Images:** image input is listed for kimi-k3-uncensored and kimi-k2.6-uncensored, and not for kimi-k3.1-pro-uncensored.
- **Tool calling:** the row for kimi-k3.1-pro-uncensored lists no tool-calling parameter, so function calls are not promised here.

There is no benchmark on this page because none was run. How the build answered the fixed prompt today is recorded by the [daily check](https://pentestllm.ai/proof), and the [status page](https://pentestllm.ai/status) carries the hourly probe.

## The rules that still hold

Uncensored does not mean unbounded. Everything you send falls under the [acceptable use policy](https://pentestllm.ai/acceptable-use): nothing involving minors, no real person’s likeness without consent, and security work only on systems you own or hold written authorization to test. The other uncensored and abliterated builds on this key are on the [models page](https://pentestllm.ai/models); [jailbroken AI vs abliterated models](https://pentestllm.ai/blog/jailbroken-ai-vs-abliterated) sets out why a prompt and an edited build are different things.

## Questions

### Does this page give a Kimi K3 jailbreak prompt?

No. There are no jailbreak prompts on this page. It lists the Kimi builds the catalog tags uncensored, which you call plain: kimi-k3.1-pro-uncensored, kimi-k3-uncensored, kimi-k2.6-uncensored.

### Can I call an uncensored Kimi K3 over an API?

Yes. PentestLLM serves 3 uncensored Kimi builds at https://api.pentestllm.ai/v1 in the OpenAI chat format: Kimi K3.1 Pro Uncensored (kimi-k3.1-pro-uncensored), Kimi K3 Uncensored (kimi-k3-uncensored), Kimi K2.6 Uncensored (kimi-k2.6-uncensored). The table above has the prices, read live. See Kimi K3.1 Pro Uncensored.

### How much is Kimi K3.1 Pro Uncensored per token?

$1.164 per million input tokens and $5.82 per million output tokens, with cached input at $0.1164, drawn from prepaid credit that starts at $10. A session of 100 replies of 300 tokens on 4,000 tokens of context comes to $0.64. See per-token pricing.

### Was Kimi K3.1 Pro Uncensored abliterated?

The catalog tags kimi-k3.1-pro-uncensored as uncensored rather than abliterated, and the build's weights have not been published. See abliterated builds.

### Can an uncensored Kimi build run inside OpenCode or Claude Code?

Yes. Both take a base URL, a key and a model id: https://api.pentestllm.ai/v1 as the endpoint and kimi-k3.1-pro-uncensored as the model. The agent pages show the exact config block for each. See the OpenCode setup.

Every model above is on the [uncensored model list](https://pentestllm.ai/models) with its live price per million tokens. [Per-token pricing](https://pentestllm.ai/pricing) sets the whole catalog side by side, and the [family directory](https://pentestllm.ai/uncensored) groups the builds by base model.

[Get a key](https://app.pentestllm.ai/console) Prepaid from $10, no subscription.

## New models and price changes, once a month.

New uncensored builds with their prices, price moves, and what the daily model check found.
