-
added 75 new models, including:
- Claude Fable 5.1;
- Claude Opus 5.5;
- GPT-6 Astra, Luna, and Sol;
- GPT-6.1 Sol;
- Gemini 3.8 Flash;
- Grok 4.7;
- DeepSeek V4.1 Flash;
- GLM 5.3 Prime;
- Qwen3.8 Max Prime;
- as well as image generation models — GPT Image 2.5, MAI-Image-2.6, and Recraft V4.1 Flash.
You can view the current list of all models and prices on selectel.ru.
-
added support for the Anthropic Messages API. Requests can now be sent using the POST /v1/messages method, and the Base URL for the Anthropic SDK is https://api.selectel.ru/aig. Supported model names include claude-haiku-4-5, claude-sonnet-4-5, and others. Claude Code, Cline, and LangChain clients work out of the box;
-
enabled automatic prompt caching on Claude models — no additional configuration is required;
-
added request cost in the API response: the usage.cost field is returned in both the OpenAI-compatible protocol and the Anthropic Messages API;
-
added the Retry-After header in the response when the rate limit is reached. It shows how many seconds to wait before retrying the request;
-
improved 400 error messages: the response now specifies which request field is invalid;
-
added the reasoning_effort parameter with the value off — it disables reasoning token generation for models that support it;
-
set a limit on the request body size — 7.5 MiB;
-
added new metrics to the detailed consumption export in CSV format:
cached_input_tokens — the number of input tokens read from the cache;
cache_write_tokens — the number of tokens written to the cache;
reasoning_tokens — the number of tokens generated by the model during the reasoning stage.