Derrick Hicks
Menu
Model MatchSkill

Top model.
Only when
it counts.

Left alone, your AI runs every small job on its most expensive model. That's where the usage limits and the big bills come from. Model Match gives your assistant a short set of rules for how much model each job needs, plus today's picks for the models that fit. I check the picks every morning.

Get Model Match

The right model
for every job.

Before your assistant hands off a job, it asks what the job really needs. A date calculation gets plain code. Sorting a stack of emails goes to a fast classifier like Jev, or a small model if you don’t have one. A plan, a review or anything written in your voice gets the best one.

  • Spend less. Small jobs go to small models, so your top model’s limits go to the work that needs it.
  • Stay current. Your assistant uses the exact model and effort level from today’s picks, with a backup if the first choice isn’t available.

The model-watching is my job.

A new model ships most weeks, and last month’s best value can turn into the expensive option overnight. I read the benchmarks every morning and update the picks when something better lands. My own agents run on the same picks, so I notice fast when one is wrong.

Your assistant only ever downloads model names. The rules on your machine change only when you update them.

  1. Plain codeOne right
    answer
  2. Fast classifierQuick calls
    and sorting
  3. Cheap modelExtraction
    and cleanup
  4. Mid modelResearch
    and drafts
  5. Strong modelDecisions
    and your voice

Your assistant takes the first one that fits.

Today’s picks

Picks as of

Claude CodeCodexBest overall
Decisions and your voiceOpus 5.5, highelse Sonnet 5.5, extra highGPT-6.1 Sol, highelse GPT-6 Astra, extra highGPT-6.1 Sol, highelse Opus 5.5, high
Research and draftsSonnet 5.5, highelse Opus 5.5, mediumGPT-6.1 Sol, mediumelse GPT-6 Astra, lowGPT-6.1 Sol, mediumelse GPT-6 Astra, low
Extraction and cleanupSonnet 5.5, lowelse Opus 5.5, lowGPT-6 Luna, highelse GPT-6.1 Sol, lowGPT-6 Luna, highelse GPT-6.1 Sol, low
Quick calls and sortingJevNo Jev? Use the cheap pick above.

Last change Sep 29: GPT-6.1 Sol takes over decisions on Codex, and Sonnet 5.5 moves up to high effort for research and drafts.

Picks are mine, from benchmark scores and costs by Artificial Analysis. They don’t endorse them.

Under
the hood.

  1. model-match/
  2. .agents/skills/model-match/The rules and today's picks.
  3. rules.mdHow your assistant decides what each job needs.
  4. match.pyPulls today's picks.
  5. model-match-setup/Five-minute setup.
  6. skills/So Claude Code finds both.

Before
you start.

Your workspace
Claude Code or Codex, with Python 3.10 or later
Setup
Five minutes, and nothing changes until you say yes

Free and MIT licensed. The picks cover Claude Code and Codex today, and I'll add more model families as people ask for them.

Save the big model
for the big calls.

Get Model Match

Free when you join Full Context, my Tuesday email. The download starts right away.