Building
Models & thinking
Choosing between 446 models, the recommended shortlist, and what the thinking level does.
Updated 21 Sept 2026
Qyant routes every request through OpenRouter, which means every chat-capable model they list is available from the picker in the builder — 446 at the time of writing, from Anthropic, OpenAI, Google, xAI, DeepSeek, Meta, Mistral, Qwen, Moonshot and others.
The picker
The model dropdown sits above the chat. Search by name, id or vendor. Each row shows the context window, the price per million tokens (input / output), and two badges: thinking (the model exposes reasoning) and images (it accepts attachments).
Your choice is remembered per browser. New accounts start on Claude Opus 5.
Recommended
The shortlist at the top is curated for building software:
| Model | Reach for it when |
|---|---|
| Claude Opus 5 | A feature has to be right the first time. Strongest at holding a large codebase in view and reasoning through refactors. The default. |
| Claude Sonnet 5 | Everyday iteration — fast enough to keep a rhythm, strong enough to carry most work. |
| GPT-6 Astra | Very large projects; questions that span many files. |
| DeepSeek V4.1 Flash | Stretching credits — small edits, copy tweaks, cleanup. |
| Qwen3.8 Max | Precise, instruction-heavy coding tasks at a low price. |
| Kimi K3 | Multi-step features from a single prompt; plans well and sticks to the plan. |
| Auto | Let the router pick per request based on the prompt. |
Everything else is under its vendor further down — older Claude versions, GPT-5 family, Gemini, Llama, Mistral and so on.
Thinking level
The Think dropdown sets how much the model reasons before answering: none, low, medium (default) or high. Higher levels produce better results on hard tasks — schema design, tricky refactors, debugging — at the cost of time and credits. For copy changes or small UI tweaks, low or none is plenty.
When the model exposes its reasoning, it streams into a collapsible Thought process card above the answer, so you can see why it made a choice.
Cost
Each request costs what the provider charges for the tokens used, counted against your plan's monthly AI allowance; one request is one credit regardless of model. Prices in the picker are per million tokens. A typical feature-sized request on a frontier model is a few cents; on the budget models it's a fraction of that.
Vision
Models with the images badge accept screenshots and mockups as attachments. If you attach an image to a model without vision support, the request is refused before it's sent.