Skip to main content
Fable 5 is gone. This is the answer you can call today. Fusion runs a panel of frontier models in parallel, then a judge returns the single strongest answer — Fable-5-level intelligence behind one slug, no waitlist. You call it like any other model; the fan-out, the judging, and the per-leg billing all happen for you.
The judge is the point. Several frontier models answer in parallel, and a judge returns the strongest answer verbatim — cross-checked against the rest, not one model’s first try, and never a black-box blend. You always get a real model’s complete answer, not a benchmark we cherry-picked.
This is the “call a preset” front door to the Routing DSL: the panels below are curated and managed for you. Want a different panel, a different judge, or your own gating rules? Compose your own in the DSL.

​
How it works

1

Route by difficulty

Casual chat goes straight to a cheap default. The hard stuff — code, agents, tool use — earns the full panel.
2

Fan out to the panel

Every model takes a swing at once — independent attempts that each catch what the others miss.
3

The judge picks the strongest

A judge reads them all and hands back the single best answer verbatim — a real model’s complete response, never a blend.
No cherry-picked benchmarks, no hidden black box: the panel and the judge are listed below, and the DSL equivalent lets you inspect every leg.

​
The panels

Three curated panels, frontier to budget. Call the slug — we keep the members current so you don’t have to.
Call it asPanelBest for
orcarouter/fusionClaude Opus 4.8 · GPT-5.5 · Gemini 3.1 ProHardest reasoning and code
orcarouter/fusion-miniClaude Opus 4.8 · GPT-5.5A leaner two-model panel
orcarouter/fusion-flashA budget panel of cheaper modelsCost-sensitive fan-out
The judge is Claude Opus 4.8 for all three.

​
Call it

Fusion is OpenAI-compatible — swap the model slug, everything else stays the same: 
curl https://api.orcarouter.ai/v1/chat/completions \
  -H "Authorization: Bearer $ORCAROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "orcarouter/fusion",
    "messages": [{"role": "user", "content": "Find the data race in this Go cache and give a corrected version."}]
  }'

​
Billing

A fan-out runs each panel member plus the judge, so a fanned-out request is billed as the sum of those underlying completions — at cost, zero markup — and only on the requests that actually fan out. Easy requests fall through to a cheaper default and bill as a single call. Each leg appears as its own row in your request logs, sharing one request.

​
Beyond the preset

The presets are best_of_n over a fixed panel. The same engine, authored as Routing DSL, hands you the whole graph:
  • Route by difficulty — fan out only on the hard requests; keep “hi” on a cheap default so you never pay panel price for chit-chat.
  • Swap the judge — pick the arbiter model, or change the strategy entirely: synthesize to fuse one answer, tests_pass to keep the patch that passes, majority, or first.
  • Build your own panel — any models, any number of legs, your own gating.
  • Cascade on a weak answer — retry with a stronger model when the first one looks low-confidence.
  • Ship it like infra — shadow a ruleset, canary it onto a slice of traffic, and dry-run it against a sample request before it serves anyone.
Load a Claude Fable 5 Level template in the DSL editor to start from a working panel and edit it to the models you use.

Compose your own panel

Pick the models, swap the judge, and set exactly when the panel fires — your whole routing graph in a few lines of YAML.