API reference

QuickstartCopy Anchor Link

Sign in to the console and create a key on the API keys page. The secret is shown once and cannot be recovered; if the key is restricted to specific models, allow the model you call.

export API_KEY="logits_live_..."  # the key you created
curl https://api.logits.com/v1/chat/completions \
  -H "Authorization: Bearer $API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "deepseek/deepseek-v4.1-flash", "messages": [{"role": "user", "content": "Hello"}]}'

These endpoints are on https://api.logits.com:

ModelsCopy Anchor Link

Text requests up to 457,728 bytes are admitted on a byte estimate; larger ones are counted with the model's own tokenizer and admitted up to 4 MiB of request body when the counted prompt plus the requested output fits the 1,048,576-token context (dense English reaches roughly one million prompt tokens within that cap; JSON string escaping counts toward it). Check the model feed for the current limits.

Supported parametersCopy Anchor Link

• temperature ranges are enforced, not clamped: a value outside the model's range returns HTTP 400 invalid_parameter.
• Passed through: the model receives the parameter, but it is not yet qualified, so the provider feed does not declare it.
• Ignored: accepted for SDK compatibility but not applied to the output.
• Best effort: passed to the model without a retry, so a forced tool call is not guaranteed.
• Reasoning efforts are none, minimal, low, medium, high, xhigh, max; none turns reasoning off. Every model also accepts include_reasoning and the OpenRouter-style reasoning: {effort, exclude} object.
• Vision: image_url parts with public HTTPS or base64 PNG/JPEG/WebP/GIF/BMP/TIFF/PPM. Source inputs fit within the 32 MiB request body (12 MiB on Responses) or the 12 MiB remote-source limit, with at most 12 MiB of each request besides its image data URLs; each prepared data URL must be under 5 MB, and prepared data URLs must total under 32 MiB. Larger sources and images above the model's max_pixels resize target are downscaled and re-encoded to fit, not rejected. Usage reports image_tokens.

Usage, errors and rate limitsCopy Anchor Link

• Usage: the OpenAI shape, with completion_tokens_details.reasoning_tokens and prompt_tokens_details.cached_tokens. Cached input is billed at the cached rate shown in the feed.
• Errors: always {"error":{"message","type","code","param"}}. Unknown models return 404 model_not_found; unknown parameters return 400 naming the parameter.
• Rate limits: a per-account, per-model rolling-minute budget of 600 requests, 8M input tokens and 2M output tokens. Responses after account/model quota admission carry x-ratelimit-limit/remaining headers for requests, input tokens and output tokens, plus x-ratelimit-reset; quota rejections add Retry-After. Validation failures and earlier rejections do not carry these headers.
• Request IDs: every response carries X-Request-Id; quote it in support requests.
• Long requests: a non-streaming request sends nothing until the whole answer is ready, and NAT gateways and firewalls on your network can drop a connection that stays idle for a few minutes (350 seconds on AWS NAT Gateway, 4 minutes by default on Azure, 10 minutes for the Google Cloud VPC firewall). The answer is then lost, often without an error. Use stream: true for requests that may run longer than a few minutes. Kimi K3.1 Pro streams keep the connection active every 15 seconds. If you must use non-streaming, enable TCP keepalive with an idle time of about 60 seconds and raise your client's request timeout.