Request parameters
Messages, max_tokens, temperature, stop, seed and other parameters.
- Can I get several answers in one request with n?No. n greater than 1 answers 400 unsupported_parameter. Send separate requests instead.FAQ
- Can message content be an array of parts?Yes, as long as every part is a text part. Image parts are refused with unsupported_content.FAQ
- Do presence_penalty and frequency_penalty work?They're accepted and passed to the model. Use small values to reduce repetition.FAQ
- Does the Model API remember earlier messages?No. Each request stands alone. Send the conversation history in messages every time.FAQ
- Does the Model API return logprobs?No. logprobs answers 400 unsupported_parameter.FAQ
- Does the Model API support JSON mode or response_format?response_format isn't among the documented parameters. Use tool calling with a schema, or ask for JSON and validate it.FAQ
- How big can a Model API request be?Up to 3 MB and 800 messages, with up to 64 tools. Larger requests answer 413 request_too_large.Reference
- How do I send a long document to SuperCosmos?Put the text in a user message. Up to 1M tokens of context, 3 MB per request; above 262K tokens input costs 1.5x.How-to
- How does max_tokens work on SuperCosmos?max_tokens caps output tokens, reasoning included. Each model has its own cap; finish_reason length means it was cut off.Explainer
- What do the finish_reason values mean?stop = finished normally, length = hit max_tokens, tool_calls = the model wants you to run a tool.Reference
- What temperature should I use with SuperCosmos?Use low temperature for code, extraction and facts; higher for brainstorming and creative writing. Range 0 to 2.FAQ
- Which message roles can I use?system, developer, user, assistant (with optional tool_calls) and tool (with tool_call_id).Reference
- Which parameters does POST /v1/chat/completions accept?model, messages, stream, stream_options, tools, tool_choice, effort, max_tokens, temperature, top_p, stop, seed and penalties.Reference
- Why are my prompt_tokens higher than my text?Every request includes a short Clatter Labs instruction block, plus message formatting, so input tokens are a bit higher.FAQ
- Will the same seed always give the same answer?Not guaranteed. seed gives best-effort repeatable sampling when the prompt and parameters are identical.FAQ
Still need help?
Tried the steps above and still stuck? Email us. We usually reply within the first hour, and always within 24 hours. A full support team is with you.
Infrastructure questions, or permanent deletion of your data
infrastructure@clatterlabs.xyzFor deleting your whole account or the data of one service.