Can it choose any model at random?

No. It only chooses from models available to your account and current API key, and the model must match the request capability.

Does it make an extra model call?

No. The system selects the model before the final request is sent. Total latency is mainly determined by the selected model.

Can I stay within one model family?

Yes. Use an exact model ID, or create different keys so each product or customer uses only the model range you intend.

What can I see after a call?

You can review the final model, usage, cost, latency, status, and request record for troubleshooting and billing review.

When should I avoid automatic modes?

Use exact model IDs for strict evals, compliance review, reproducible experiments, or customer-specified models.

One API, a suitable model automatically

Stop guessing which model to use. Put smart routing in the model field.

Call the OpenAI-compatible API as usual. Simple tasks can use faster or lower-cost models, harder requests can use stronger models, and code or vision workloads can use dedicated routes. Fixed model IDs still work whenever you need them.

Get API Key Read technical docs

Change one BaseURL

Match the task automatically

Review cost and latency

One request, automatic selection

auto/balanced

curl https://api.chuizi.ai/v1/chat/completions \

-H "Authorization: Bearer $CHUIZI_API_KEY" \

-H "Content-Type: application/json" \

-d '{"model":"auto/balanced","messages":[{"role":"user","content":"..."}]}'

How to call it

model: auto/balanced

Best for

Text, code, and vision tasks

After the call

Usage, cost, and latency are visible

Choose a routing mode by task

You do not need to memorize every model. Use auto/balanced for normal production traffic. Switch to a clear model name when you need speed, lower cost, stronger reasoning, code, or vision.

auto/balanced

Default for production

Balances quality, speed, and cost for most application requests.

A practical first step when moving from fixed model IDs to automatic selection.

auto/fast

Low-latency first

Prioritizes faster models for support, formatting, extraction, and short answers.

Use it when user experience depends more on response time than maximum reasoning depth.

auto/cheap

Cost first

Keeps classification, summarization, and structured output on lower-cost capable models.

Good for batch processing and high-volume automation.

auto/quality

Quality first

Prioritizes stronger models for complex reasoning, analysis, and long-context work.

Use it when answer quality matters more than the lowest possible cost.

auto/code

Code and agents

Built for coding, debugging, tool calls, code review, and agent workflows.

Keeps engineering requests on routes that fit code work better.

auto/vision

Vision understanding

Built for images, screenshots, document understanding, and visual QA.

Image-input requests are not sent to text-only models.

You can still pin a model: use explicit IDs such as openai/gpt-4.1 or qwen/qwen3.6-plus. Automatic modes and fixed models can live in the same project.

The user experience should stay simple

Your application sends the request. Chuizi.AI matches a model, handles temporary unavailability, and records usage without pushing that complexity into your business code.

Send the request as usual

Keep one BaseURL and the OpenAI-compatible format.

Understand the task

Detect whether the request needs speed, low cost, code, vision, or deeper reasoning.

Match a suitable model

Choose a working route from the models available to the current account.

Review the result

See the model, usage, cost, latency, and status after completion.

Move model choice out of application code

Do not maintain model rules and cost checks in every app. Your product calls one stable API while the selection strategy can evolve with your workload and model availability.

Fixed models and automatic modes can coexist: use exact model IDs for evals, auto/balanced for normal production traffic, and dedicated modes for code or vision.

Fixed models still work

Use exact model IDs for evals, compliance, or customer-specified models.

Separate scenarios cleanly

Production, testing, support, agent, and media workloads can use different keys and model ranges.

More stable when paths fail

If one path is temporarily unavailable, the system prefers a working path to reduce disruption.

Cost is visible per request

Every request can show usage, cost, status, and latency for review.

Which model value should I use?

You do not need to choose from every model every time. Pick a stable model value for the scenario.

Need	Suggested model	Best for
A specific model is required	openai/gpt-4.1	Evals, compliance, customer-specified models
Normal production requests	auto/balanced	Most text and mixed business traffic
Realtime UX or support	auto/fast	Short answers, classification, extraction, formatting
High-volume low-cost tasks	auto/cheap	Batch summaries, labels, structured output
Code and agent workflows	auto/code	Generation, debugging, review, tool calls
Images or document understanding	auto/vision	Screenshots, PDFs, visual QA

FAQ

No. You can keep using exact model IDs. Smart Routing only runs when model is set to an automatic mode such as auto/balanced, auto/code, or auto/vision.

Start with auto/balanced and let the system handle model choice.

Create a key and call immediately. Use fixed models when results must be exact, and automatic modes when you want efficiency and resilience.

Get API Key View model catalog

Exact availability, pricing, and capabilities depend on the model catalog and current API key settings in the console.