Mythical Pro Agents

The headline frontier models — the ones marked Pro — are slow, expensive, and gated behind separate consumer subscriptions or per-token bills. Oracle is the single CLI that talks to all of them with the same flags, the same session store, and the same bundling rules.

#The lineup

Plus any OpenRouter id — e.g. minimax/minimax-m2, openai/gpt-4o-mini, qwen/qwen-2.5-coder-32b-instruct — when you set OPENROUTER_API_KEY.

#When to reach for which

#GPT-5.5 Pro (default)

The current "Oracle of last resort." Slow (10 minutes typical, hour+ for huge bundles), expensive on API, free if you have ChatGPT Pro. Best for: long-form architecture review, multi-file refactors, gnarly bugs that need to chew on the whole repo.

oracle --engine browser --model gpt-5.5-pro \
  -p "Plan the auth migration end-to-end" \
  --file "src/auth/**" --file "docs/auth.md"

#Gemini 3 Pro

Free in browser mode if you're signed into gemini.google.com in Chrome. Also the path for YouTube transcript analysis (--youtube) and image generation/edit (--generate-image, --edit-image).

oracle --engine browser --model gemini-3-pro \
  --prompt "a minimalist eye-of-providence logo, vector" \
  --generate-image out.png --aspect 1:1

#Claude Opus 4.1

API-only. Best at sustained reasoning over a single tightly-scoped problem; lower hallucination rate on tricky code than Pro GPT models in our experience. Use when the task is "reason carefully" rather than "write a lot."

oracle -p "Audit this state machine for unreachable transitions" \
  --model claude-4.1-opus --file "src/state/**"

#Multi-model triangulation

When the answer matters, ask three:

oracle -p "Will this migration corrupt user data under concurrent writes?" \
  --models gpt-5.5-pro,gemini-3.1-pro,claude-4.1-opus \
  --file "migrations/0042_user_schema.sql" \
  --file "src/db/**"

Oracle aggregates per-model cost in the run summary. See Multi-model.

#Deep Research (browser only)

Browser mode can flip ChatGPT into Deep Research mode for cited reports:

oracle --engine browser --model gpt-5.5-pro \
  --browser-research deep \
  -p "Survey approaches for embedded vector search in a Rust app, with citations"

The captured report lands at ~/.oracle/sessions/<id>/artifacts/deep-research-report.md. Deep Research can run for an hour or more — Oracle stores the session and reattaches.

#Thinking time

Pro / Thinking models in browser mode accept a --browser-thinking-time knob:

oracle --engine browser --model gpt-5.5-pro \
  --browser-thinking-time heavy \
  -p "Refactor this hot path" --file "src/render/**"

#Pricing notes (rough heuristics, not contracts)

• GPT-5.x Pro (API): tokens cost meaningfully more than non-Pro. Watch the run summary.
• GPT-5.x Pro (browser): "free" with ChatGPT Pro / Plus subscription, but slow.
• Gemini 3 Pro (browser): free with a Google account.
• Claude Opus 4.1: per-token API only.
• OpenRouter ids: pricing varies wildly per provider; always preview with --dry-run summary.

--files-report plus --dry-run summary is the right reflex before any Pro run on a large bundle. Token counts ≠ dollars, but they're a close-enough proxy.

#Engine compatibility

See provider-specific docs for the gory details: OpenAI / Azure / OpenRouter, Gemini, Anthropic, Grok.