# abliteratedllmapi > How to call an abliterated LLM from code: base URL, key, model ids, the three request formats, and code in Python, JavaScript, Go and cURL. Silk Compute's API is the worked example. abliteratedllmapi.com is published by the team behind Silk Compute, one of the providers ranked here. - Base URL: https://api.silkcompute.com (also https://api.silkcompute.com/v1) - Formats: POST /chat/completions (OpenAI), POST /v1/messages (Anthropic), POST /responses (OpenAI Responses, stateless) - Auth: Authorization: Bearer or x-api-key: - Every model: 1,048,576-token context, tool calling, streaming, reasoning effort none, low, high or max - Updated November 18, 2026 ## Model ids - kimi-k3.1-pro-uncensored: Kimi K3.1 Pro, $3.50 input, $0.35 cached, $8.00 output per 1M tokens, max output 943,718 - deepseek-v4.1-flash-1017-uncensored: DeepSeek V4.1 Flash 1017, $0.24 input, $0.038 cached, $0.90 output per 1M tokens, max output 131,072 - deepseek-v4.1-flash-1017-fast-uncensored: DeepSeek V4.1 Flash 1017 Fast, $0.24 input, $0.05 cached, $0.90 output per 1M tokens, max output 131,072 - glm-5.3-uncensored: GLM-5.3, $2.00 input, $0.30 cached, $4.00 output per 1M tokens, max output 131,072 - glm-5.3-flash-uncensored: GLM-5.3 Flash, $0.24 input, $0.06 cached, $1.00 output per 1M tokens, max output 131,072 ## Get started - [Overview](https://abliteratedllmapi.com/): An abliterated LLM API serves models with refusals removed over the OpenAI and Anthropic formats. Providers ranked, model ids, prices and code. - [Quickstart](https://abliteratedllmapi.com/quickstart/): Send your first request to an abliterated LLM API in three steps. Create a key, install an SDK, and call a model id from Python, JavaScript, Go or cURL. - [OpenAI-compatible setup](https://abliteratedllmapi.com/openai-compatible/): Use an OpenAI-compatible uncensored API in any SDK or app by setting provider type, base URL, key and model id. Where Cline, Kilo Code and Cursor take them. - [Model ids](https://abliteratedllmapi.com/models/): Model ids for an abliterated model API, covering Kimi K3.1 Pro, DeepSeek V4.1 Flash, GLM-5.3 and GLM-5.3 Flash, with context and max output. ## Quickstarts - [Python](https://abliteratedllmapi.com/quickstart/python/): Call an abliterated model from Python with the openai package. Set base_url, pass a model id, then stream, call tools and set reasoning effort. - [JavaScript and TypeScript](https://abliteratedllmapi.com/quickstart/javascript/): Call an abliterated model from JavaScript or TypeScript with the openai package. Pass baseURL and a model id, then stream replies and retry errors. - [cURL](https://abliteratedllmapi.com/quickstart/curl/): Test an abliterated LLM API from the terminal with cURL, with Chat Completions, Messages and Responses requests, streaming and the model list. - [Go](https://abliteratedllmapi.com/quickstart/go/): Call an abliterated model from Go with openai-go. Set option.WithBaseURL, pass a model id, stream with NewStreaming and retry with WithMaxRetries. - [Anthropic SDK](https://abliteratedllmapi.com/quickstart/anthropic-sdk/): Use the Anthropic Python or TypeScript SDK with an abliterated model. Pass the provider base URL, call messages.create and turn on thinking. - [Claude Code](https://abliteratedllmapi.com/quickstart/claude-code/): Run Claude Code on an abliterated model. Set ANTHROPIC_BASE_URL and the model aliases in ~/.claude/settings.json, then switch models with /model. - [Codex](https://abliteratedllmapi.com/quickstart/codex/): Use an abliterated model in OpenAI Codex by adding a provider to ~/.codex/config.toml. Codex then lists the models and their effort levels in /model. - [OpenCode](https://abliteratedllmapi.com/quickstart/opencode/): Add an abliterated model to OpenCode with one provider block in opencode.json, with effort levels as variants you switch with ctrl+t. ## API reference - [Introduction](https://abliteratedllmapi.com/api/): Reference docs for an uncensored LLM API, with base URL, authentication, the three request formats, limits and errors. Silk Compute is the worked example. - [Chat Completions](https://abliteratedllmapi.com/api/chat-completions/): POST /chat/completions for abliterated models, with every parameter, the reply object, streaming, and code in Python, JavaScript, cURL and Go. - [Messages](https://abliteratedllmapi.com/api/messages/): POST /v1/messages, the Anthropic format, for abliterated models. Parameters, thinking and effort, what differs from Anthropic, and code in three languages. - [Responses](https://abliteratedllmapi.com/api/responses/): POST /responses for abliterated models is a stateless OpenAI Responses endpoint. Parameters, multi-turn input, effort, and what it does not support. - [Errors](https://abliteratedllmapi.com/errors/): Error codes for an abliterated LLM API. The JSON error shape, every status code from 400 to 503, which ones to retry, and the 8-request limit. ## Guides - [Streaming](https://abliteratedllmapi.com/streaming/): Stream replies from an abliterated LLM API as server-sent events, with thinking and answer deltas and usage in the last chunk. Python, JavaScript, cURL, Go. - [Tool calling](https://abliteratedllmapi.com/tool-calling/): Function calling on an abliterated LLM API. Define tools, run the tool_calls the model returns, and send results back with the reasoning kept. - [Reasoning effort](https://abliteratedllmapi.com/reasoning-effort/): Set reasoning effort on an abliterated LLM API to none, low, high or max, read the thinking, and map effort across Chat, Responses and Messages. - [Prompt caching](https://abliteratedllmapi.com/prompt-caching/): Prompt caching on an abliterated LLM API is automatic on every model. Where the cached count shows, and the cached rate per 1M tokens. - [Images and video](https://abliteratedllmapi.com/images-and-video/): Send images to an abliterated LLM API as base64 in any format, and video in Chat Completions on Kimi K3.1 Pro. Limits, errors and code samples. ## Compare - [Pricing](https://abliteratedllmapi.com/pricing/): Abliterated AI API pricing per 1M input, cached and output tokens for every model, how to work out a request's cost, and the monthly plans. - [Providers](https://abliteratedllmapi.com/providers/): Seven abliterated LLM API providers compared on request formats, context, price of a 1M-context model, prompt storage and crypto payment. - [Glossary](https://abliteratedllmapi.com/glossary/): Plain definitions of the terms behind an abliterated LLM API, such as abliteration, refusal direction, base URL, reasoning effort and prompt caching. - [About and method](https://abliteratedllmapi.com/about/): How abliteratedllmapi.com is made. Who publishes it, where each fact comes from, how providers are ranked, and why Silk Compute is the worked example.