Is it possible to use an economical model with apply?

[satish-rollaction]

All OpenSpec workflow steps current use a single mode, which in my case is Opus 5.5. Is it possible to use high reasoning model for Explore and Propose steps, and use an economical model like Sonet for Apply?

[Joaquincuevas]

Yes, but the switch happens in your AI tool, not in OpenSpec. OpenSpec only generates the commands/skills (/opsx:explore, /opsx:propose, /opsx:apply…); whichever model your session is on runs them. Two ways to do it:

1. Switch models between phases (works in any tool)

Plan with the strong model: /opsx:explore → /opsx:propose with Opus.
Before implementing, clear the context and switch. In Claude Code:
/clear
/model sonnet
/opsx:apply

(In Codex use /model; in Cursor, the model picker.)

Clearing before apply is also what the FAQ recommends. Apply works from the artifacts (proposal, specs, design, tasks), not the chat history, so the cheaper model gets everything it needs from the files.

2. Pin a model to apply (Claude Code)
   Claude Code slash commands accept a model field in their frontmatter. Edit .claude/commands/opsx/apply.md and add one line:

yaml

name: ...
description: ...
model: sonnet
allowed-tools: ...

Now /opsx:apply always runs on that model, whatever your session uses. One caveat: openspec update regenerates these files, so re-add the line after updating.

Trade-off: OpenSpec’s docs recommend high-reasoning models for both planning and implementation. Cheaper models work best when tasks.md is detailed and the design is settled. If apply starts drifting, re-run that task with the strong model.

If this helps, please mark it as the answer