DocsConcepts
Needs
Tasks, purposes and constraints: what the model must do.
A need says what the model must do. Give a task or a purpose, plus any of these constraints:
| Parameter | What it does |
|---|---|
task | What the model does.One of text.chat, text.reasoning, code, vision.understand, vision.extract, image.generate, image.edit, embedding, rerank, speech.transcribe, speech.synthesize, video.generate. |
purpose | A reviewed recipe named after what an app does; it sets the task, needs and usage (see purposes). |
needs | Capabilities the model must have on your provider, optionally with a level, e.g. input.pdf:native,tools (see capabilities). |
budget | A band relative to the task (see budget bands), or a hard max_price per typical call in USD.One of cheap, medium, premium. |
optimize | cost (the default) takes the cheapest model within quality_margin (5%) of the best; quality the best within the budget; value the best score per unit of cost.One of cost, quality, value. |
origin | The model maker's country, e.g. us, eu or not:cn. It follows the model, not the host. |
input_tokens, output_tokens, images … | Usage: sets the typical call the prices are compared on. Defaults come from the purpose, else 1,000 input and 300 output tokens. |
min_quality, min_context … | min_quality, min_context, released_after, allow_preview, exclude, license, service_tier and currency narrow it further. |
max_memory_gb, max_vram_gb, max_disk_gb | The device's limits, for local runtimes. |
The answer's need echoes the need as resolved: your parameters merged with the purpose and the defaults.