RedRouter FAQ

RedRouter is a low-refusal inference API for authorized security work, including penetration testing, red teaming, bug bounty research and defensive code review. It serves modified RR models and standard open-weight models through OpenAI-compatible and Anthropic-compatible endpoints. Choose an explicit model ID for each request; RedRouter does not silently substitute a different model.

RedRouter offers refusal-reduced open-weight models through an API, often described as an uncensored model API. RR models are modified to reduce refusals, and RedRouter does not add a refusal layer of its own. Models can still decline requests or produce incorrect output, and the Terms of Service still apply.

Abliterated models are models modified to reduce refusal behavior, and RedRouter labels its refusal-reduced variants RR. The RR label does not establish an exact weight revision or guarantee a particular result. Check each model’s catalog details for reported checkpoint facts and fields marked unknown.

RedRouter is for developers and security teams doing work they are authorized to perform. Examples include in-scope penetration tests, red-team engagements, bug bounty research, malware analysis and review of code you own or have permission to assess. Keep targets and rules of engagement explicit, and review model output before using it.

RedRouter supports OpenAI-compatible Chat Completions and Responses APIs and an Anthropic-compatible Messages API. For OpenAI SDKs, set the base URL to https://redrouter.org/v1; for Anthropic clients, use https://redrouter.org so the client can append /v1/messages. The APIs support text, streaming and tool calling, with documented field and model limits.

RedRouter provides setup instructions for Claude Code, Cline and OpenCode using a RedRouter API key and an explicit model ID. Claude Code uses the Messages endpoint; Cline and OpenCode can use the OpenAI-compatible endpoint. Compatibility depends on the client’s requested features: these are text models, and unsupported fields or hosted tools have documented limitations.

RedRouter’s RR models are GLM 5.3 Flash RR (glm-5.3-flash-rr), GLM 5.3 RR (glm-5.3-rr, experimental), Qwen 3.8 27B RR (qwen3.8-27b-rr) and DeepSeek V4.1 Flash RR (deepseek-v4.1-flash-rr). Standard variants are GLM 5.3 Flash, GLM 5.3, Qwen 3.8 27B and DeepSeek V4.1 Flash releases 1017 and 0910. Start with GLM 5.3 Flash RR for a refusal-reduced model with tool calling and a 1,048,576-token context; check Models for current availability and limits.

GLM 5.3 Flash RR costs $0.20 per million input tokens, $0.08 per million reported cached input tokens and $0.80 per million output tokens at standard rates. RedRouter uses prepaid credits; unreported cache reads use the input rate. Optional Developer ($20), Growth ($50) and Scale ($200) monthly card plans include an equal amount of credit, which expires at period end and is spent before non-expiring top-ups.

RedRouter retains request logs by default; prompt and completion storage depends on your key settings and opt-in Zero Data Retention. New API keys have content storage off, existing keys start with storage on, and browser Chat and the Mac app store content unless account Zero Data Retention overrides it. Stored content is encrypted for up to 30 days for abuse prevention, support, debugging and billing disputes, with restricted, audited access; usage metadata is still recorded when content storage is off.

Turn on Zero Data Retention in RedRouter Settings to stop storing prompts and completions for subsequent requests across your API keys and Chat. Zero Data Retention adds 20% to every token rate on pay-as-you-go and is included during paid monthly plan periods. It disables Save chat, while usage metadata remains available for billing; it does not independently verify retention by the infrastructure running the model.

RedRouter permits penetration testing and red teaming only on systems you own or are authorized to test. The Terms of Service prohibit unauthorized attacks or access, illegal activity, harassment and interference with the service. A low-refusal model does not grant permission to test a target or override an engagement’s scope.

RedRouter provides a programmable inference API for applications, scripts and coding agents, alongside browser Chat and a Mac app. You select a model ID, authenticate with an API key and pay from prepaid credits. Chat is available when you want a conversation; the API is available when you want to connect your own tooling.

RedRouter accepts prepaid credit top-ups by card or USDC and offers optional monthly card plans. The minimum top-up is $10, and top-up credits do not expire. Add credits from the Credits page after creating an account; credits are non-transferable and have no cash value.

Create a RedRouter account, create an API key, add credits and set REDROUTER_API_KEY in your environment before sending a request. This three-line curl example calls GLM 5.3 Flash RR through Chat Completions with a small output limit.

curl --fail-with-body https://redrouter.org/v1/chat/completions \
  -H "Authorization: Bearer $REDROUTER_API_KEY" -H "Content-Type: application/json" \
  -d '{"model":"glm-5.3-flash-rr","max_tokens":256,"messages":[{"role":"user","content":"List three checks for an authorized code security review."}]}'