# Silk Compute

> Silk Compute is an uncensored LLM API. It serves abliterated open-weight models (Kimi K3.1 Pro, Hy4, DeepSeek V4.1 Flash 1017, GLM-5.3 and GLM-5.3 Flash) with their refusal behavior removed and no content filters added. The API is compatible with OpenAI Chat Completions, Anthropic Messages and OpenAI Responses, so the official SDKs, Claude Code, Codex and OpenCode work by changing the base URL and API key.

- API base URL: https://api.silkcompute.com
- Chat route: POST /chat/completions (OpenAI format). Also POST /v1/messages (Anthropic format) and POST /responses (OpenAI Responses, stateless).
- Auth: an API key from https://platform.silkcompute.com, sent as a Bearer token or x-api-key header.
- Billing: monthly plans with a usage allowance, from $10, or prepaid credits from $5, by card or crypto (Monero, Bitcoin, Ether, USDT, USDC). Charged per token.
- Privacy: prompts and model outputs are not stored. Only billing records such as token counts and cost are kept.

## Models

Prices are USD per 1M tokens. Every model has a 1,048,576-token context window and supports tool calling and streaming.

- [Kimi K3.1 Pro Uncensored](https://silkcompute.com/models/kimi-k3.1-pro-uncensored): model id `kimi-k3.1-pro-uncensored`. For the hardest workloads: agentic coding, research and long outputs, with reasoning effort low, high or max. Takes text, images and video. $5.00 input, $0.50 cached input, $15.00 output. Max output 943,718 tokens.
- [Hy4 Uncensored](https://silkcompute.com/models/hy4-uncensored): model id `hy4-uncensored`. Tencent's newest flagship, with reasoning effort low, high or max. Takes text and images. $1.60 input, $0.30 cached input, $3.20 output. Max output 65,536 tokens.
- [DeepSeek V4.1 Flash 1017 Uncensored](https://silkcompute.com/models/deepseek-v4.1-flash-1017-uncensored): model id `deepseek-v4.1-flash-1017-uncensored`. Fast reasoning with effort low, high or max. Takes text and images. $0.28 input, $0.07 cached input, $1.00 output. Max output 131,072 tokens.
- [GLM-5.3 Uncensored](https://silkcompute.com/models/glm-5.3-uncensored): model id `glm-5.3-uncensored`. The full GLM-5.3, with reasoning effort low, high or max. Text only. $2.00 input, $0.30 cached input, $4.00 output. Max output 131,072 tokens.
- [GLM-5.3 Flash Uncensored](https://silkcompute.com/models/glm-5.3-flash-uncensored): model id `glm-5.3-flash-uncensored`. Lowest cost, with reasoning effort low, high or max. Takes text and images. $0.24 input, $0.06 cached input, $1.00 output. Max output 131,072 tokens.

## Docs

- [Quickstart](https://silkcompute.com/docs/quickstart): create a key, add credits, send a first request in Python, JavaScript or cURL.
- [Models](https://silkcompute.com/docs/models): model ids, context windows, output limits and reasoning effort.
- [Reasoning](https://silkcompute.com/docs/reasoning): set thinking effort and read the model's reasoning.
- [Streaming](https://silkcompute.com/docs/streaming): stream replies with usage.
- [Tool calling](https://silkcompute.com/docs/tool-calling): function calling in the OpenAI and Anthropic formats.
- [Prompt caching](https://silkcompute.com/docs/prompt-caching): the cached input rate and how to get more cache hits.
- [Anthropic and Responses APIs](https://silkcompute.com/docs/anthropic-and-responses): what the compatible endpoints support.
- [Errors](https://silkcompute.com/docs/errors): error codes and when to retry.
- [Billing and credits](https://silkcompute.com/docs/billing): top-ups, payment methods and per-token charges.

## Coding agents

- [Integrations](https://silkcompute.com/docs/integrations): every setup guide on one page.
- [Claude Code](https://silkcompute.com/docs/claude-code): settings.json to run Claude Code on these models.
- [Codex](https://silkcompute.com/docs/codex): config.toml provider setup for Codex.
- [OpenCode](https://silkcompute.com/docs/opencode): opencode.json provider setup.
- [Cursor](https://silkcompute.com/docs/cursor): your key and base URL in Cursor's settings.
- [Hermes Agent](https://silkcompute.com/docs/hermes): config.yaml provider setup.
- [DeepSeek Harness](https://silkcompute.com/docs/deepseek-harness): cordis.patch.yml provider setup.
- [Other tools](https://silkcompute.com/docs/other-tools): Kilo Code, Cline, Roo Code, Open WebUI and other OpenAI-compatible tools.

## Enterprise

- [Enterprise](https://silkcompute.com/docs/enterprise): what an Enterprise agreement covers and how to start one.
- [Single sign-on](https://silkcompute.com/docs/sso): SAML SSO, SCIM and roles on an Enterprise agreement.
- [Security and compliance](https://silkcompute.com/docs/security): data protection, compliance status and documents on request.
- [HIPAA](https://silkcompute.com/docs/hipaa): talk to us before sending protected health information.

## Guides

- [Security research](https://silkcompute.com/docs/security-research): malware analysis, reverse engineering and detection engineering.
- [Red teaming](https://silkcompute.com/docs/red-teaming): authorized penetration testing and attack path planning.
- [Trust and safety](https://silkcompute.com/docs/trust-and-safety): classifier training data and moderation testing.

## Blog

- [Abliterated AI: a complete guide to choosing a model](https://silkcompute.com/blog/what-is-an-abliterated-model): what abliteration is, a size chart from laptop to rack, which model suits which job, and what each costs in a coding agent.
- [Best uncensored models for coding agents](https://silkcompute.com/blog/uncensored-model-for-coding-agents): what an agent needs, which model suits which job, and what each costs at an agent mix.
- [Venice AI API alternative](https://silkcompute.com/blog/venice-ai-api-alternative): Silk Compute and the Venice API compared.
- [abliteration.ai alternative](https://silkcompute.com/blog/abliteration-ai-alternative): Silk Compute and abliteration.ai compared on price and plans.
- [Uncensored models on OpenRouter](https://silkcompute.com/blog/openrouter-uncensored-models): whether OpenRouter has abliterated models, and why routing does not remove refusals.
- [Kimi K3.1 Pro Uncensored: specs, pricing and best uses](https://silkcompute.com/blog/kimi-k3-1-uncensored): image and video input, 1M tokens of context, up to 131K output tokens, and what it's best for.
- [Hy4 Uncensored: specs, pricing and best uses](https://silkcompute.com/blog/hy4-uncensored): image input, benchmarks against the other models here, and which jobs to give it.
- [DeepSeek V4.1 Flash 1017 Uncensored is live](https://silkcompute.com/blog/deepseek-v4-1-flash-uncensored)
- [GLM-5.3 Flash Uncensored is live](https://silkcompute.com/blog/glm-5-3-flash-uncensored)

## Optional

- [Models and pricing](https://silkcompute.com/models): the full price table.
- [Privacy policy](https://silkcompute.com/privacy)
- [Terms of use](https://silkcompute.com/terms)
- [Contact](https://silkcompute.com/contact): hello@silkcompute.com