Skip navigation
Claude Code tip: keep Opus 5.5 as your main model, but stop paying Opus prices for your subagents
0:00 / 0:32

Claude Code tip: keep Opus 5.5 as your main model, but stop paying Opus prices for your subagents

delost

@thedelost5.1K followers

Visit on X
146 likesDownloadOriginal post

28,222 views

Claude Code tip: keep Opus 5.5 as your main model, but stop paying Opus prices for your subagents move them to Sonnet 5.5 Opus 5.5 plans and decides Sonnet 5.5 subagents do the heavy reading, editing and testing at half the price ($2 / $10 vs $4 / $20 per 1M tokens) Fable 5.1 stays on call with /advisor and only speaks up at three points: → before a plan: is this the right approach? → when the same error comes back: am I digging in the wrong place? → before "done": what did I miss? Opus thinks. Sonnet does. Fable checks Jev engineering is the same move one layer down: the forks that need no thinker (which file, which tool, retry or stop) go to Jev in under half a second, and the big models only see the ones that split - the full tree > Opus 5.5 on high runs the main session > explorer reads the code > worker edits and runs tests > researcher pulls the docs > all three on Sonnet 5.5, medium > Fable 5.1 on call as the advisor paste the tree and this prompt into Claude Code ↓ "Rebuild my Claude Code setup around this tree: 1. Check ~/.claude/agents and .claude/agents for subagents that already fit explorer, worker and researcher. > Draft new ones only for missing roles > Give each model: claude-sonnet-5-5, effort: medium > List any that pin a different model before changing them 2. Keep the main session on Opus, set effortLevel to high in ~/.claude/settings.json and advisorModel to fable 3. Find anything that keeps the advisor off (CLAUDE_CODE_DISABLE_ADVISOR_TOOL, DISABLE_TELEMETRY, any variable that stops feature-flag fetching) plus CLAUDE_CODE_EFFORT_LEVEL, which overrides subagent effort. Report them, change nothing 4. Add one rule to ~/.claude/CLAUDE.md: consult the advisor before a large plan, when an error repeats, and before calling a long task done Show me every change as a diff first. No edits until I say go." ↳

8 Comments

Comments from the original post on X (Twitter)

  • LiveRichMedia's profile picture

    @JayRichMusic

    the core setup is good for substantial coding work. I would adjust it before making it your everyday default. • Keep the separate roles. Having agents investigate, implement, and review is sensible. Codex officially supports this arrangement. Subagent documentation • Question Sol high on every task. OpenAI recommends comparing settings on your actual work and keeping the lightest setting that meets your quality standard. Extra agents also consume additional tokens. Your Terra medium default remains consistent with your cost rules. Model guidance • Keep Astra available for difficult reviews. Give it a specific question and the actual changes, errors, and test results. Calling it automatically because a task was long can produce unnecessary review. Its agreement still does not prove the work functions. • Understand auto_review correctly. It reviews eligible permission requests. It does not check every action or certify finished work. Configuration documentation My recommendation: adopt the role separation and occasional independent review; retain your routine default and approval rules for stronger models.

    1 like

  • David Chen's profile picture

    @diousk00

    gpt-6-luna is the right one as a subagent

    1 like

  • Will's profile picture

    @FixChatWill

    opus for the plan sonnet for the loop is how you keep the bill from eating the win

  • catman's profile picture

    @catmanyau

    Think of it as a team: Opus is the lead, Sonnet handles the work, and the advisor is the reviewer at key checkpoints.

  • The AI Maximalist's profile picture

    @theaimaximalist

    Opus leads the strategy and Sonnet does the work it knows best which feels very efficient to watch in action

  • Dima T.'s profile picture

    @Dimentorio

    the real optimization is routing each task to the model that actually needs to handle it

  • Ignition of Motivation's profile picture

    @MititelSergey

    Gemini 4 Argon (announced Sept 30): → 1M-token output (was 64K) → Leads on 12 of the 18 benchmarks Google shared (trails on 2 coding ones) → $2 in / $10 out per 1M tokens (intro price, $4/$20 later) → Built for coding, long multi-step work, cyber defense

  • Fox's profile picture

    @foxnet2000

    别的不说,这个动效做的是相当棒!