A field guide to picking the right model for the job

“Which model is best?” is the wrong question. The right one is “best at what?” Here’s the mental map our team uses every day.

1. Hard thinking: frontier reasoning models

For multi-step problems — architecture decisions, tricky bugs, strategy, maths — reach for a frontier model with extended reasoning. They’re slower and pricier per message, but they plan before they answer and are far less likely to confidently skip a step.

Tell-tale sign you need one: you’d want a senior colleague to sleep on it.

2. Everyday work: fast models

Rewriting an email, summarising a thread, turning notes into a list. Fast models handle the bulk of daily tasks in a fraction of the time. Using a frontier model here is like taking a jet to the corner shop.

3. Big documents: long-context models

Contracts, research papers, codebases, transcripts. Some models comfortably read hundreds of pages in one go and can point to exactly where they found something. Ask for page references — and spot-check one or two.

4. Code: write with one, review with another

Coding-tuned models are excellent at producing working code quickly. A different model reviewing the diff catches a surprising number of issues the author missed — the same reason human teams do code review.

5. High volume: open-weight models

Open-weight families like Llama, Mistral, Qwen and DeepSeek are strong, fast and increasingly competitive. They shine for high-volume, repetitive tasks and make a great second opinion next to closed models.

6. Pictures and motion: image and video models

Image models differ wildly in style. Some excel at photorealism, others at typography or illustration. Run the same prompt across two or three in Compare before committing to a direction.

When in doubt

Use Auto. It routes each message to a sensible model and always shows you which one answered — a quick way to build your own intuition over time.