Ox Alpha: the anonymous model that was in the Yenisei catalog

Update, September 3, 2026: the model has been withdrawn from the catalog. Its place is taken by Yenisei Max (yenisei-max).

In late August, Ox Alpha briefly appeared in the catalog. It was an anonymous release: one of the leading labs was testing a model in public without revealing its name. This happens often: strong models are published under a neutral name before the official announcement so they can be tried on real requests.

What the model was

• Reasoning by default. Thinking was a mandatory part of the work; its depth could be adjusted, but it could not be turned off.
• Code and long agent workflows. Chains of dozens of steps, tool calls and edits, where it matters not to lose the thread by the end of the chain.
• 1M-token context. A codebase, a long conversation and logs fit in whole.
• Text, images and video as input.
• Tool calling and structured output for agents.

What to switch to

• Yenisei Max (yenisei-max) is the direct replacement: reasoning, image and video input, a million tokens of context.
• Kimi K2.7 Code (kimi-k2.7-code) if you need a model built specifically for code and development agents.
• GLM 5.3 (glm-5.3) for long agent chains and a context window larger than a million tokens.

curl https://api.yenisei.ru/v1/chat/completions \
  -H "Authorization: Bearer $YENISEI_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "yenisei-max",
    "messages": [
      {"role": "user", "content": "Find the data race in this snippet and suggest a fix."}
    ]
  }'

What the alpha story teaches

An anonymous release is a trial run. The author can change the model, limit the load, and at some point withdraw it entirely, which is exactly what happened here. So do try new models, but do not tie your workflows to a single model without a fallback. With a unified API, a fallback model is one line in your settings.

Current prices are on the plans and pricing page.