Gemini 2.5 Pro and 3.7 Flash: Google's multimodality and what to use instead in Yenisei

Gemini 2.5 Pro and Gemini 3.7 Flash are Google models that accept not only text and images but also audio and video. Both have a 1M-token context.

There are no Gemini models in the Yenisei catalog, but we do have open-weight models that work with more than text. Below is which ones.

Gemini 2.5 Pro and 3.7 Flash

Pro is the top model: text, images, audio, and video in one request, step-by-step reasoning, long context. It is used when the task is complex and the data is mixed: video analysis with conclusions, analytics over mixed materials.

Flash is a fast model with the same set of inputs, optimized for speed and price.

What to use in Yenisei

• MiMo v2.5 (mimo-v2.5): text, images, audio, and video as input, reasoning, and a context of about a million tokens. Closest to Gemini in its feature set.
• Qwen3.8 Flash (qwen-flash): text, images, and video, fast and inexpensive; the Flash counterpart.
• Kimi K3 (kimi-k3): a top model with images and video as input and a million-token window; the Pro counterpart.
• Gemma 4 31B (gemma-4-31b): Google's own open model: images and video as input, reasoning.
• Yenisei Max (yenisei-max): our top model with images and video.

For audio-only tasks, such as speech recognition and analyzing conversations, there is a dedicated Voxtral Small 24B (voxtral-small).

Why a model that accepts audio and video

Many models read text and images but not audio or video. If your product works with sound and video, a model that accepts them directly is often more practical than a separate "speech recognition plus text model" pipeline.

Example: text plus image

curl https://api.yenisei.ru/v1/chat/completions \
  -H "Authorization: Bearer $YENISEI_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "mimo-v2.5",
    "messages": [{
      "role": "user",
      "content": [
        {"type": "text", "text": "Describe what is wrong with this chart and suggest how to fix it."},
        {"type": "image_url", "image_url": {"url": "https://example.com/chart.png"}}
      ]
    }]
  }'

How to choose

• A complex task with images or video: Kimi K3, MiMo v2.5, or Yenisei Max.
• A stream with images and video where speed and price matter more: Qwen3.8 Flash or Gemma 4 26B (gemma-4-26b).
• A purely text task: you don't need images or sound, so take a text model such as DeepSeek V4 Pro (deepseek-pro).

One OpenAI-compatible API, payment in rubles, 1 credit = 1 ₽.