> ## Documentation Index
> Fetch the complete documentation index at: https://apione.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Choose models by use case

> Compare quality, latency, context, capabilities, stability, and cost for your workload.

APIONE should not recommend one fixed model for every user. Define the task, then compare currently available models with representative business samples.

![APIONE model selector with capabilities and input/output prices.](https://mintcdn.com/apione/pAFvFGkbmcRZirTZ/images/ui/model-selector.png?fit=max\&auto=format\&n=pAFvFGkbmcRZirTZ\&q=85\&s=5ab2c2637860f5f187a680268134fabf)

APIONE model selector with capabilities and input/output prices.

## Answer five questions first

1. Is the input pure text, images, audio, or video?
2. Do you prioritize accuracy, speed, or cost?
3. How much context and output length do you need?
4. Will you use tool calling, JSON, image understanding, or code editing?
5. Can your business accept switching models when one is temporarily unavailable?

## Scenario dimensions

| Scenario | Compare first |
| - | - |
| Coding and agents | Tool use, long context, patch quality, time to first byte |
| Support and summarization | Format stability, language quality, batch cost, latency |
| Complex reasoning | Accuracy, inference time, maximum output, failure rate |
| Vision | Image count, resolution, OCR, chart understanding |
| Image generation | Prompt adherence, text rendering, sizes, editing capabilities |
| Video generation | Duration, aspect ratio, motion consistency, task time and cost |

## Do not rely only on leaderboards

Public leaderboards only reflect specific test sets. Actual business performance is also affected by prompts, protocol, tool definitions, context length, routing, and parameters. Prepare at least 10 to 30 sanitized business samples, and compare them under uniform prompts and scoring criteria.

## Build a small evaluation table

| Metric | How to record |
| - | - |
| Task success rate | Whether it meets explicit acceptance criteria |
| Time to first byte | Time from request to first output |
| Total time | Time until the complete result is returned |
| Input / output usage | Response `usage` and console details |
| Cost per call | Actual charge under the selected group |
| Format pass rate | Whether JSON, tool calls, or templates are parseable |
| Stability | Failures and variance across repeated runs on the same sample |

## Recommended strategy

* Primary model: meets the quality floor and covers the main capabilities.
* Fast model: used for classification, rewriting, simple Q\&A, and preprocessing.
* High-quality model: used only for complex tasks or escalation after failure.
* Media model: selected separately by frame size, duration, output format, and cost.

Model IDs, prices, and availability change. The application should fetch the current list from `/v1/models` and parameters from `/v1/models/{model}/params`; do not hard-code recommended model IDs into core business logic.

Next: [Model parameters](https://apione.ai/docs/en/models/parameters.md), [Pricing groups](https://apione.ai/docs/en/account/pricing-groups.md), [Request reliability](https://apione.ai/docs/en/api/request-reliability.md).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.