Changelog

New models and capability changes. Anything that can alter a response you already depend on lands here.

Added glm5.3

  • Open-weight mixture-of-experts reasoning and coding model with a 262k context window.

Added glm5.3-flash

  • Open-weight 321B-parameter (18B active) reasoning model with a 262k context window.

deepseek4-flash capability declarations

  • Tool calling, JSON mode, structured outputs, logprobs and reasoning are now declared in the feed.
  • logprobs is the only response field unique to this model.

Added qwen3.8-27b

  • Open-weight general-purpose model with a 262k context window and native tool calling.

inference.api.reka.ai is live

  • An OpenAI-compatible endpoint for the open models Reka serves on its own GPUs.
  • Opening roster: deepseek4-flash, glm5.2, reka-flash-3 and reka-edge-2603.
  • deepseek4-flash and glm5.2 serve a 262,144-token window with 131,072 tokens of output.
  • Video, image and audio models are callable through the same key.