Ask a panel. Get one answer.
Ask one AI and you get one opinion — with that model's blind spots baked in. Fusion asks a panel of the world's best models the same question at the same time, then a judge reads every answer, spots where they agree, where they conflict, and what each one missed — and writes you a single answer you can trust more than any one model alone.
Your question fans out to several frontier-class models at once — like specialists consulting on one case. Identities stay masked.
- ✓ circles the consensus
- ✕ flags contradictions
- ◌ notes the blind spots
One clean answer backed by the whole panel — a second, third, and fourth opinion, delivered as one. Billed in credits like any other request.
When to use Fusion, what it costs, and how to call it
When to use it
Hard questions where being right beats being cheap: research, architecture calls, legal-ish reviews, gnarly debugging. For routine traffic, normal routing is the better deal — Fusion is the escalation, not the default.
What it costs
A panel costs panel prices — published, like everything here: 20 credits per ~1k tokens on Pro (advanced panel), 120 on Pro Max (frontier panel), billed from the same wallet as everything else.
How to call it
Send model: "intelliverse/fusion" on any Pro or Pro Max key. Panel composition is tier-managed; panelist identities are masked in response metadata like any other request. Enterprise security reviews get the full panel-composition and routing policy — see Enterprise.
One worked example
const res = await client.chat.completions.create({
model: "intelliverse/fusion", // panel + judge, one request
messages: [{ role: "user", content:
"Should we shard this Postgres table or partition it? 40M rows, 3k writes/min." }],
});
// One synthesized answer. A ~1k-token exchange bills 20 credits on Pro
// (advanced panel) or 120 on Pro Max (frontier panel).Latency: the panel answers in parallel, then the judge synthesizes — budget roughly two sequential model calls end-to-end. Escalate to Fusion when correctness matters more than milliseconds.
Fusion FAQ
How much slower is Fusion than a normal request?
A panel plus a judge is inherently slower than a single model — expect a response in roughly the time of two sequential model calls (the panel runs in parallel; the judge reads it afterwards). Use it where being right beats being fast; keep routine traffic on normal routing.
What does a Fusion call cost?
Published panel prices, billed from the same wallet as everything else: 20 credits per ~1k tokens on Pro (advanced panel), 120 on Pro Max (frontier panel).
Is this the llm-council pattern?
Same idea — several strong models answer, review each other, and a chairman synthesizes — offered as a hosted, metered API instead of a repo you run and maintain yourself. Send one model string; the panel, judging, and billing are handled for you.
Try it live
The playground below streams a real Fusion request through your own API key.
Run a panel of models side-by-side, judge the results, and fuse the best answer.
Start free to run this — no cardSecond opinions, on tap
Start free to explore the platform, then unlock Fusion on Pro or Pro Max.