[FEATURE]: Implement support for TOON to reduce token usage and inference costs #10773

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

Hi OpenCode team 👋

First of all, great work on the project. We’ve been using OpenCode in real scenarios and it’s shaping up really nicely.

I’d like to propose adding support for TOON (Token Optimized Output Notation) as an optional output mode, with the explicit goal of reducing token usage and inference costs, especially in agent-based and iterative workflows.

Why TOON?

In many use cases (agents, tools, structured responses, follow-up calls), verbose natural language output is unnecessary and expensive. TOON allows the model to:

Produce highly compact, structured outputs

Minimize repetition and filler tokens

Improve latency and cost efficiency

Maintain semantic clarity while being machine-friendly

This is particularly valuable when:

Running multi-step agents

Chaining model calls

Operating under tight token or budget constraints

Using LLMs as reasoning engines rather than chat interfaces

Proposed Implementation

Some ideas on how this could be integrated (totally open to discussion):

Add a configurable output mode, e.g.:

outputFormat: "natural" | "toon"

Allow TOON to be enabled:

Per request

Per agent

Globally via config

Provide a minimal TOON schema or guideline so outputs stay predictable

Expected Benefits

📉 Lower token consumption

💸 Reduced inference costs

⚡ Faster responses

🤖 Better fit for agentic and automation-heavy workflows

Final Thoughts

This wouldn’t replace natural language outputs, but rather complement them for performance-critical paths. In my opinion, TOON support would be a strong differentiator for OpenCode compared to more chat-oriented tools.

Happy to help with:

Spec definition

Examples

Testing / validation

Thanks for considering it 🚀