oh-my-opencode.md

Suggested omo Optimization

IMPORTANT: Right now, GTP-5.3-codex is not available via Copilot. It is probably about to change soon, and might replace many GPT-5.2 uses. Additionally, the Sonnet-4.6 multiplier might be subject to change - according to Github.

Goal

The prompt to Claude Sonnet-4.6 was:

Check all guides in here: https://github.com/code-yeongyu/oh-my-opencode/tree/dev/docs/guide. I need some advice which model to use under which circumstances. Note that I have a Copilot subscripton with limited requests, and premium request multipliers listed here: https://docs.github.com/en/copilot/concepts/billing/copilot-requests#model-multipliers. After considering all this information: Have a look at ~\.config\opencode\oh-my-opencode.json. Write an optimized file (oh-my-opencode_optimized.json) in the same folder, regarding also the requests pricing. All models listed in the pricing table are available.

Result

Agent/Category	Original	Optimized	Why
sisyphus	claude-opus-4.6 max (3×)	claude-sonnet-4.6 (1×)	Sisyphus is the orchestrator — it reads files, delegates, routes. It doesn't need Opus reasoning. Sonnet 4.6 is the current generation equivalent of what you're running today as the main agent (it's the model powering your session right now). Saves 3× on every user interaction.
oracle	gpt-5.2 high (1×)	gpt-5.2 high (1×)	Keep — Oracle is the expensive consultant, 1× is already the best you can do for high-IQ reasoning without going to Opus.
librarian	claude-sonnet-4.5 (1×)	gpt-4.1 (0×/free)	Librarian does web searches and doc lookups. GPT-4.1 is free with Copilot and perfectly capable for retrieval tasks.
explore	gpt-5-mini (0×/free)	gpt-5-mini (0×)	Already optimal.
multimodal-looker	gemini-3-flash-preview (0.33×)	gemini-3-flash-preview (0.33×)	Keep — still the best multimodal option at low cost.
prometheus	claude-opus-4.6 max (3×)	claude-sonnet-4.6 (1×)	Prometheus plans/interviews — structured text generation, not deep reasoning. Sonnet is more than sufficient and runs at 1/3 the cost. Alternative: gpt-4.1 (0×/free) and switch to Sonnet after the interview part.
metis	claude-opus-4.6 max (3×)	gpt-5.2 (1×)	Metis does pre-planning gap analysis — analytical/structured. GPT-5.2 is a strong analyst at 1× vs 3×.
momus	gpt-5.2 medium (1×)	gpt-5.2 medium (1×)	Keep — already good choice for plan review.
atlas	claude-sonnet-4.5 (1×)	claude-sonnet-4.6 (1×)	Updated to current gen.
visual-engineering	gemini-3.1-pro-preview high (1×)	gemini-3.1-pro-preview (1×, no variant)	Removing variant: high — variants mean extended thinking tokens. For UI/CSS tasks that's wasteful.
ultrabrain	gemini-3.1-pro-preview high (1×)	gpt-5.2 high (1×)	Ultrabrain = hard logic/reasoning. GPT-5.2 is purpose-built for this and fits better than Gemini.
artistry	gemini-3.1-pro-preview high (1×)	claude-sonnet-4.6 (1×)	Same cost, but Claude's "narrative reasoning" family is better suited for creative/unconventional approaches per the agent-model-matching guide.
quick	claude-haiku-4.5 (0.33×)	gpt-4.1 (0×/free)	GPT-4.1 is free AND more capable than Haiku for trivial tasks.
unspecified-low	claude-sonnet-4.5 (1×)	gpt-4.1 (0×/free)	Free model for low-effort tasks — no reason to spend 1× here.
unspecified-high	claude-sonnet-4.5 (1×)	claude-sonnet-4.6 (1×)	Updated to current gen, same cost.

This will lead to this oh-my-opencode.json:

{
  "$schema": "https://raw.githubusercontent.com/code-yeongyu/oh-my-opencode/master/assets/oh-my-opencode.schema.json",
  "agents": {
    "sisyphus": {
      "model": "github-copilot/claude-sonnet-4.6"
    },
    "oracle": {
      "model": "github-copilot/gpt-5.2",
      "variant": "high"
    },
    "librarian": {
      "model": "github-copilot/gpt-4.1"
    },
    "explore": {
      "model": "github-copilot/gpt-5-mini"
    },
    "multimodal-looker": {
      "model": "github-copilot/gemini-3-flash-preview"
    },
    "prometheus": {
      "model": "github-copilot/claude-sonnet-4.6"
    },
    "metis": {
      "model": "github-copilot/gpt-5.2"
    },
    "momus": {
      "model": "github-copilot/gpt-5.2",
      "variant": "medium"
    },
    "atlas": {
      "model": "github-copilot/claude-sonnet-4.6"
    }
  },
  "categories": {
    "visual-engineering": {
      "model": "github-copilot/gemini-3.1-pro-preview"
    },
    "ultrabrain": {
      "model": "github-copilot/gpt-5.2",
      "variant": "high"
    },
    "artistry": {
      "model": "github-copilot/claude-sonnet-4.6"
    },
    "deep": {
      "model": "github-copilot/claude-sonnet-4.6"
    },
    "quick": {
      "model": "github-copilot/gpt-4.1"
    },
    "unspecified-low": {
      "model": "github-copilot/gpt-4.1"
    },
    "unspecified-high": {
      "model": "github-copilot/claude-sonnet-4.6"
    },
    "writing": {
      "model": "github-copilot/gemini-3-flash-preview"
    }
  }
}