Help·Chat and models 4 min

Which model for what?

The model families and their strengths — for text, code, research, large documents and images.

The studio offers models from all major providers. ’Auto’ chooses for each task itself — if you want to choose specifically, here are the rules of thumb. The concrete versions change every few months, the strengths of the families remain surprisingly stable. As of: September 2026.

  1. 1

    Claude (Anthropic)

    First choice for demanding texts, analyses and programming: careful, precise, good tone, sticks to long instructions and project knowledge. Opus is the strongest tier for difficult tasks and agents, Sonnet the everyday standard, Haiku the fast, cheap variant for simple tasks. Also available as EU version via Frankfurt.

  2. 2

    ChatGPT / GPT (OpenAI)

    The all-rounder: very broadly usable, strong with tools, web search, tables and documents, very large context windows for long materials. The Pro tier thinks longer and delivers the most thorough answers, the fast variants (Mini, Luna) are cheap for routine.

  3. 3

    Gemini (Google)

    Strong with very long documents and mixed content (text, images, tables in one), good summaries, very fast Flash variants at a small price. Also as EU version via Frankfurt. The image models of the family are called Nano Banana.

  4. 4

    Grok (xAI)

    Current affairs, social media and casual tone: Grok has access to live data from X and is often most up-to-date on daily topics. For factual documentation, rather Claude or GPT.

  5. 5

    Mistral and EU models

    Mistral from France is the choice when data should stay in Europe; Large for normal tasks, Small for fast and cheap. In EU mode (Settings) you only see models with the 🇪🇺 badge — Claude and Gemini via Frankfurt, Mistral from Paris.

  6. 6

    DeepSeek, Qwen, Kimi, GLM, MiniMax

    Very good value for money for routine: summarising, translating, rephrasing, simple code. Note with sensitive data: Chinese providers, partly via US servers — with Safeguard or in EU mode, better the European options.

  7. 7

    Images: photorealistic

    Seedream (ByteDance) and the Nano Banana models deliver the most realistic photos, product images and portraits. Seedream Pro for marketing images with text, Nano Banana Pro for dense instructions and edits on existing images.

  8. 8

    Images: graphics, text in image, editing

    GPT Image understands instructions most accurately and writes text in the image reliably — ideal for posters, logos with lettering, infographics. Nano Banana Pro and GPT Image are also the right choice for marking and targeted changes.

  9. 9

    Images: illustration and versatility

    Flux 2 Pro is the all-rounder for illustration, art and design styles; Flux Dev and Schnell are cheap variants for drafts. Imagen (Google) is available as EU version for images that should be generated in Europe.

  10. 10

    Cheap or thorough?

    For routine, the fast models are enough (Haiku, Luna, Flash, Mistral Small) — a fraction of the price. For concepts, analyses and code, the strong tier (Opus, GPT Pro) with thinking depth ’Thorough’ is worth it. Unsure? Multi-query: two models in parallel, then compare.

Tips

  • ’Auto’ knows these rules and chooses them per message — including image model when you describe one.
  • Prices per model are under Settings → Models & Prices; the cost of each answer under the answer.
  • Too many models in the menu? There also ’Let AI suggest selection’ — the AI keeps the currently best ones.

Open Models & Prices

Related guides

Tidy up model selection

Only the currently best models in the menu — via AI suggestion or manually.

Multi-query: compare models

Up to four models answer in parallel, plus one merged response.

EU models and data protection

Which models run in Europe, what EU-only mode does and where the data processing agreement is.