Skip to main content
This page lists new models, retirements of outdated models and price changes. We announce retirements in advance: here, in our Telegram channel, in a console announcement, and by email to everyone who has used the model in the last 30 days. Current prices are always on the Pricing page. This log starts in October 2026.

​
Upcoming

​
Retiring deepseek-v4-flash
DeepSeek has retired this version — requests to the old name are already served by the new Flash. On 1 November 2026 the name deepseek-v4-flash will be turned off on RuAPI, and requests using it will return an error. Replacement: deepseek-flash, the same model at the same price. Change the model name in your client settings or code.
​
Retirement of claude-sonnet-4-5-20250929
Anthropic is ending support for Claude Sonnet 4.5 (source). On 30 November 2026, claude-sonnet-4-5-20250929 will be retired. Its replacement is claude-sonnet-5-5 — just change the model name in your client settings or code.
​
gemini-4-argon price change
Google’s current price for gemini-4-argon is a limited-time offer that ends on 30 November 2026 (source). From 1 December 2026, Google doubles the price, and RuAPI’s input, output and cache-read prices for gemini-4-argon double too. See the pricing page for current prices.

​
Changes

​
DeepSeek Flash: new version and the deepseek-flash name
New name deepseek-flash. On 17 October 2026 DeepSeek released a new version of Flash, DeepSeek-V4.1-Flash-1017 (source); in DeepSeek’s API the model is called deepseek-flash. On RuAPI the name is available at the same price as deepseek-v4.1-flash, and requests to it are served by the new version. Use deepseek-flash in new projects.What improved, according to DeepSeek:
  • better at finishing whole projects from scratch, such as games and 3D scenes;
  • spends less time and fewer output tokens on tasks, so they cost less;
  • specifically optimized for DeepSeek Harness — if you use it, update to version 0.2.1-rc.1 or later.
Image input. deepseek-flash accepts images, diagrams and screenshots — we tested it in both the OpenAI and the Claude format; deepseek-v4-pro is text-only. More in Image input.The earlier name deepseek-v4.1-flash keeps working at the same price — it is served by the new version too, so there’s no need to change your settings. The name deepseek-v4-flash will be turned off on 1 November 2026 (see “Upcoming” above) — replace it with deepseek-flash.
​
Gemini 4 Argon
New gemini-4-argon: Google’s flagship reasoning model. About 1.05M tokens of context and up to 65,536 output tokens per response; supports image input, function calling and streaming. The current price is Google’s limited-time offer and doubles on 1 December 2026 (see “Upcoming” above). More in the Gemini models overview.
​
New Qwen models and pricing updates
New Alibaba Qwen models: qwen3.8-max, qwen3.8-max-0902, qwen3.8-flash, qwen3.8-27b, qwen3.7-max, qwen3.7-plus. For qwen3.7-plus, requests with more than 256K input tokens are billed at a higher rate, as on Alibaba Cloud.gemini-3.1-flash-lite-preview retired: Google shut down this preview model on 25 May 2026. For lightweight tasks, use gemini-3.5-flash-lite or gemini-2.5-flash-lite.Pricing aligned with Google’s rules:
  • gemini-2.5-pro and gemini-3.1-pro-preview: requests with more than 200K input tokens are billed at a higher rate, as on Google. Shorter requests are unchanged.
  • gemini-2.5-flash-lite: the cache-read price now matches Google’s official price.
Full prices are on the Pricing page.
​
Claude Haiku 5.5
New model: claude-haiku-5-5, a fast, low-cost Claude model. As on Anthropic, the price depends on prompt length: requests with more than 100K input tokens are billed at a higher rate. See the Claude models overview for details.