[FEATURE]: Cost-aware multi-provider routing with free-model and credit optimization #52782

Feature hasn't been suggested before.

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

• I have searched the existing issues and confirmed this feature has not already been requested in the same form.

Description

As a follow-up to #52743, OpenCode's future Auto routing mode should not only choose the appropriate model and reasoning level, but also optimize which provider or endpoint should execute the request.

Many users have access to multiple providers at the same time:

• OpenRouter
• OpenAI
• Anthropic
• Google
• Groq
• Together AI
• Fireworks AI
• DeepInfra
• other OpenAI-compatible providers
• local/self-hosted models

Some providers may also offer free models, prepaid credits, subscriptions, promotional quotas, or cheaper inference for the same/similar models.

OpenCode could use this information when routing requests.

Example

Instead of:

Task → selected model

the routing process could become:

Task → required capabilities → candidate models → candidate providers → cost/quota/availability → best route

For example:

• simple task → eligible free model
• normal coding task → inexpensive open model
• complex task → stronger paid model
• free quota exhausted → cheapest suitable paid route
• subscription quota available → prefer subscription before API spend
• provider unavailable/rate-limited → automatically use an allowed fallback
• difficult task or failed attempt → escalate to a stronger model

OpenRouter example

OpenRouter is particularly interesting because it provides many models and free endpoints behind one API.

If a user has OpenRouter configured, OpenCode could optionally prefer eligible free models first and only consume paid credits when the task requires it.

The same concept should not be limited to OpenRouter. Direct providers or local models could be preferred whenever they offer a better route.

User policy

Users should control the optimization strategy rather than manually selecting a model for every task.

For example:

Auto – Quality
Auto – Balanced
Auto – Save
Auto – Free first

A custom policy could eventually allow preferences such as:

• prefer free models
• prefer existing subscriptions/quotas
• minimize API spending
• prefer local models
• preserve premium quota
• prioritize quality
• maximum cost per request/session
• allowed/blocked providers

Manual model selection should always remain available.

Transparency

OpenCode should clearly show why a route was selected.

Example:

Qwen → OpenRouter Free
Reason: sufficient capability + free quota available

or:

Claude → Anthropic
Reason: complex task + subscription quota available

The user should ideally see the estimated cost before or after execution where pricing information is available.

Relationship to #52743

#52743 proposes a provider-agnostic intelligent model and compute router.

This issue extends that idea specifically with economic routing:

model quality + task requirements + provider + price + free quota + subscription quota + credits + availability

The goal is not simply to always choose the cheapest model.

The goal is to use the cheapest route that is sufficiently capable for the task, while respecting the user's quality and provider preferences.

Long term, this could make:

/models → Auto

a true intelligent routing layer instead of only an automatic model selector.