log in

changelog

Changes, week by week.

Customer-visible changes to the API, model catalog, pricing, status page and docs. Newest first.

  • Kimi K3.1 Pro is available: $2.00 input, $0.20 cached input and $10.00 output per million tokens, with a 1,048,576-token context. Retired kimi-k3 IDs return model_not_found.
  • Structured output accepts unions and recursive schemas for DeepSeek and Kimi.
  • Sign-up no longer needs an invitation.
  • The docs have a sidebar, search and a page table of contents.
  • The docs list the temperature range per model: 0-2 for DeepSeek V4.1 Flash, 0-1 for Kimi K3.1 Pro.
  • The pricing page links the JSON price list at https://api.logits.com/models.

  • /v1/messages accepts Anthropic-format requests from Claude Code with DeepSeek V4.1 Flash, including images in user and tool-result content.
  • status.logits.com publishes per-model health, uptime, TTFT, duration and throughput.
  • The pricing page lists context and max output per model. Credits expire 12 months after purchase or grant.
  • GET /v1/models with an API key lists only the models that key may call.
  • Chat Completions accepts replayed histories with Anthropic cache_control hints or tool calls without type.
  • /v1/responses reasons by default and fills requested reasoning summaries. It accepts store: true but stores nothing, and drops a top-level hosted web_search tool instead of rejecting it.
  • Structured output accepts allOf inheritance for DeepSeek and Kimi.

  • /v1/responses serves DeepSeek V4.1 Flash and Kimi K3 natively, with reasoning, the apply_patch tool and images.
  • Codex, including the Codex engine in ChatGPT desktop, connects to /v1/responses. Setup is at /docs/codex.
  • Chat Completions accepts WebP and GIF images in addition to PNG and JPEG.
  • Contexts are 1,048,576 tokens, and request bodies up to 4 MiB are accepted.
  • Model IDs with a vendor prefix, such as deepseek/deepseek-v4.1-flash, are accepted.