One identity. A private model mixture.

BrokenGPT routes each request through a mixture of open-source models. The individual checkpoints and providers are private implementation details and may change as quality, speed, and availability improve.

Public contract

API alias: broken-one
Product identity: broken-one
Advertised context cap: 1,048,576 tokens
Runtime: open-source model mixture online

Clients always receive the stable BrokenGPT alias. Requests may be routed to different members of the mixture, but upstream model and provider identifiers are neither added to responses nor exposed by the public API.

No gateway censorship

BrokenGPT does not hard-block prompts with a keyword moderation filter. The router selects an inference tier for capability and availability; every prompt is sent to the configured mixture.

Accuracy and limits

Outputs can be false, biased, incomplete, outdated, or insecure. Verify important factual claims and generated code, especially before medical, legal, financial, safety-critical, or high-impact use.

Data flow

Web-chat messages are stored to provide conversation history. Direct /v1/chat/completions requests are not stored as conversation content by the application; usage metadata such as token counts, latency, key ID, request status, and the public alias is recorded for metering and operations.

Open-source mixture

Operators are responsible for verifying the licenses and commercial terms of every model used in the private mixture. BrokenGPT describes the system at the product level without turning backend checkpoint names into a public promise.