Skip to main content
​
​

Product

  • Chat rooms can now run audio, moderation and embeddings models as well as text models (#201, #157).
  • Foundry, a new home for experimental projects, opens with LLM Council, which sends one prompt to several models and merges their answers into one (#201).

API & SDKs

  • Go, C#, Java, PHP and Ruby SDKs are now available alongside the TypeScript and Python SDKs, all at version 1.1.0 (#203, #201).
  • The SDKs 1.1.0 warn when a request names a deprecated or retired model, and their model types and route paths match the API (#201, #203).
​

Product

  • Promo codes can be redeemed for credits on the new Redeem page, linked from Settings -> Credits; teams billed by invoice cannot redeem them (#188).
  • Sign-up no longer fails when an optional account setup step errors (#187).
​

Product

  • Chat shows the model name above each assistant message instead of the model ID, and the name links to the model’s page (#179).
  • On phones, the header has a single menu with navigation, account links such as settings, credits and keys, and a light, dark or system theme switch (#180, #183, #181).
​

Product

  • The gateway quickstart on model pages offers only the model IDs that the selected endpoint and its active providers accept (#157).
​
​
​

Product

  • Settings -> Provisioning Keys is now Settings -> Management API Keys, and the old address redirects to the new page (#153).
  • Header search ranks model name matches first, so searching Grok 4.20 puts Grok 4.20 at the top (#156).
  • New guide: SpaceXAI Grok 4.20 covers reasoning and multi-agent effort controls when adopting Grok 4.20 (#155).

API & SDKs

  • Breaking: The /v1/provisioning/keys aliases are removed; call /v1/management/keys instead; see Management API Keys (#153).
​

Models

Aion 2.5 model release artwork
​

Product

  • Model pricing tables show capabilities a provider has turned off, such as batch, as Not Active instead of hiding them (#141).

API & SDKs

  • Requests accept service_tier set to standard or priority again; see Service Tiers (#141).
  • The openai/gpt-latest alias now points to openai/gpt-5.4 (#140).
​

API & SDKs

  • The openai/gpt-chat-latest alias now points to openai/gpt-5.3-chat (#138).
​

API & SDKs

  • The Responses API accepts and returns the assistant message phase (commentary or final_answer), so multi-turn requests can send it back; see Create a response (#130).
  • Responses requests to OpenAI models accept provider_options.openai.context_management with type: "compaction" and an optional compact_threshold; see Parameters (#130).
  • Requests to OpenAI that omit safety_identifier now send one taken from user or user_id, or else the request ID (#130).
Last modified on October 28, 2026