# bf16.ai > bf16.ai is an uncensored LLM API. One key and one prepaid USD balance reach > uncensored open-weight models — refusal behaviour removed from the weights, > not papered over with a prompt — through OpenAI-compatible, OpenAI Responses > and Anthropic Messages endpoints. "bf16" is a brand name, not a statement > about numeric precision. ## Connect a tool - OpenAI-compatible base URL: `https://api.bf16.ai/v1` (Chat Completions and Responses). - Anthropic Messages base URL: `https://api.bf16.ai` — **no `/v1`**. Clients that speak the Anthropic protocol, Claude Code among them, append the path themselves; adding `/v1` yourself is the most common cause of a 404 here. - Auth: `Authorization: Bearer sk-bf16-v1-...`, or `x-api-key` on the Anthropic endpoint. Keys start with `sk-bf16-v1-`. - List what is live: `GET https://api.bf16.ai/v1/models` (public models without a key; pass `Authorization: Bearer sk-bf16-v1-...` to verify a key and include any account-allowlisted models). - Create a key at https://bf16.ai/dashboard/keys. The console shows a ready-to-run `curl` and Claude Code command with the key already filled in, and you can read your own key back later unless you created it without a recoverable copy. ## Client setup guides - [All client setup guides](https://bf16.ai/docs/clients) - [Claude Code](https://bf16.ai/docs/clients/claude-code): Anthropic’s terminal coding agent, pointed at bf16 instead of Anthropic. (anthropic-messages) - [Codex](https://bf16.ai/docs/clients/codex): OpenAI’s coding agent, on a separate bf16 profile. (openai-responses) - [OpenCode](https://bf16.ai/docs/clients/opencode): Open-source terminal agent with a provider registry. (openai-completions) - [Pi](https://bf16.ai/docs/clients/pi): Terminal agent that speaks the Anthropic Messages format. (anthropic-messages) - [OpenClaw](https://bf16.ai/docs/clients/openclaw): Personal agent with a non-interactive onboarding flow. (openai-completions) - [Hermes](https://bf16.ai/docs/clients/hermes): Personal agent configured from the command line. (openai-completions) - [OpenMausBot](https://bf16.ai/docs/clients/openmausbot): Third-party bot that speaks the OpenAI-compatible format. (openai-completions) - [curl and the SDKs](https://bf16.ai/docs/clients/curl): The shortest proof the key works, and the two SDK shapes. (openai-completions) ## Models - `deepseek/deepseek-v4.1-flash-uncensored` — DeepSeek V4.1 Flash (Uncensored). $0.26/$1 per 1M in/out, 1M context, 130K max output, text + image. - `zai-org/glm-5.3-flash-uncensored` — GLM-5.3 Flash Vision (Uncensored). $0.3/$1.12 per 1M in/out, 1M context, 131K max output, text + image. - `zai-org/glm-5.3-uncensored` — GLM-5.3 Flagship (Uncensored). $1.35/$4.2 per 1M in/out, 1M context, 131K max output, text. - `deepseek/deepseek-v4.1-flash-uncensored-fast` — DeepSeek V4.1 Flash (Uncensored Fast). $0.28/$1.12 per 1M in/out, 1M context, 130K max output, text + image. - `moonshotai/kimi-k3.1-pro-uncensored` — Kimi K3.1 Pro (Uncensored). $2.9/$9.9 per 1M in/out, 1M context, 131K max output, text + image. Prices are USD per 1M tokens, live from the catalog. Cached input is billed at a lower rate; see https://bf16.ai/pricing. ## Rate limits Limits are per account, across every key and model, and follow lifetime net top-ups. A request over the limit is refused before tokenization with a 429 carrying `concurrency_limit_exceeded`; retry once one of your in-flight requests finishes. - L0 (Free): 3 concurrent, 180 requests/minute. Reached at $0 of lifetime top-ups. - L1 (Tier 1): 10 concurrent, 600 requests/minute. Reached at $10 of lifetime top-ups. - L2 (Tier 2): 30 concurrent, 1800 requests/minute. Reached at $100 of lifetime top-ups. - L3 (Tier 3): 50 concurrent, 3000 requests/minute. Reached at $500 of lifetime top-ups. - L4 (Tier 4): 100 concurrent, 6000 requests/minute. Reached at $2000 of lifetime top-ups. ## Billing - Prepaid. You top up a USD balance and each request is deducted at the model's token rates. Purchased credit does not expire. - One-time top-ups from $10 to $2000, whole dollars, by card through a merchant of record. There is no subscription and no off-session auto-recharge. - A low-balance email alert is configurable in the console. ## Questions and answers - What does "uncensored" mean on bf16.ai? The refusal direction has been edited out of an open-weight model's weights. No retraining is involved. Refusals are reduced, not guaranteed gone, and an upstream provider may still block a clearly illegal request. - Is there a free trial? A small credit is granted automatically once your email address is verified. See https://bf16.ai/pricing. - What is off-limits? Illegal use. See https://bf16.ai/legal/aup. - Do you store prompts? No. See https://bf16.ai/legal/privacy. ## Core pages - [Models](https://bf16.ai/models): the live catalog with per-model detail. - [Pricing](https://bf16.ai/pricing): token rates and the rate-limit tiers. - [Docs](https://bf16.ai/docs): endpoints, headers and budget controls. - [Cookbook](https://bf16.ai/cookbook): worked recipes. - [Playground](https://bf16.ai/playground): try a model in the browser. - [Status](https://bf16.ai/status): gateway readiness. - [Blog](https://bf16.ai/blog) - [Acceptable Use](https://bf16.ai/legal/aup) · [Privacy](https://bf16.ai/legal/privacy) · [Terms](https://bf16.ai/legal/terms) · [Refunds](https://bf16.ai/legal/refunds) ## Optional - [Full reference](https://bf16.ai/llms-full.txt) - [Sitemap](https://bf16.ai/sitemap.xml) Not affiliated with Anthropic, OpenAI, Google, Z.ai, Moonshot AI or DeepSeek. Model names are the trademarks of their respective owners.