Stop paying flagship prices for easy questions.
Fugal is a research demo of a simple idea: don't send every question to the most expensive model. A tiny 0.6B router reads your question, predicts the odds each of 17 models would get it right, and calls the one that earns it — once. No verifier, no retries, no second call. You see the predicted odds and the real price of every choice.
How a question flows
One question, two steps. That's the whole pipeline.
The demo's default is Route mode: step 1 laid bare — the full ranked board, every model's predicted odds and estimated price — always free, no model called. Answer (sign-in) sends the question on to the routed model and streams the reply.
The numbers — and what they measured
Honesty first: the archived benchmark below measured an earlier research harness — a four-step cascade (route → answer → verify → escalate) whose verifier and escalation are no longer part of the product. The numbers hold for that system, so that is how they're labeled:
Even for the cascade, the honest reading was parity accuracy at roughly half the cost — its +0.9 pt edge is not statistically significant (McNemar p=1.0); the 2.1× cost win is what survives the statistics. The product now ships the route-once path, which hasn't been measured end-to-end yet. Two goalposts bound it on the same 110 questions: a fixed cheap model scored 0.946 at $0.0024/query and the full cascade 0.973 at $0.0079 — and the verifier was the net that caught the router's mistakes. Until the route-once number lands, the claim here is the mechanism and the board, not an accuracy headline. The cascade-era breakdown, the Pareto frontier, and every domain score are on the Research scoreboard.
Routers rot — so the scoreboard refreshes
Model prices and capabilities shift almost monthly. A router trained on last month's prices slowly makes worse choices — it rots. Rather than hide that, Fugal treats the leaderboard as something to re-measure on a cadence: re-price the models, re-score new ones, recalibrate the router, publish a dated snapshot. A leaderboard that visibly decays and refreshes is the artifact almost nobody publishes — and it's exactly what a neutral, ongoing benchmark needs to be.
Where this is heading: a decentralized benchmark roadmap
The same refresh loop — fresh, un-memorized tasks scored by a deterministic grader — is the seed of a Bittensor subnet: miners submit routing/answering systems, a reference validator scores them each epoch on burn-once tasks, and the frontier scoreboard becomes the public, continuously-re-measured record of who's actually on the cost/accuracy frontier. Fugal ships as the open, fork-me reference miner. This part is design-stage, not live yet — but it's why the honest, re-gradeable numbers matter from day one.
Honest by design
Every number here traces to an archived, re-gradeable artifact —
gate_matrix_2026-07.json and e2e_cache_2026-07.json, re-scored
offline by figs/emit_frontier.py — and when a number describes a system we no
longer ship, it says so instead of quietly borrowing the credit. Ties and failures are
reported plainly; corrections are logged rather than overwritten. It's a research demo, so
answers come from third-party models via OpenRouter, sign-in is optional (guest chats live
only in your browser; an account adds cloud sync), and it's rate-limited.
FAQ
Is this free?
Route mode is always free — the router runs one local forward pass and no model is called, so nothing is spent. Answer mode is free to you too, but it spends the demo's OpenRouter budget, so it's sign-in gated and capped per day.
What's the difference between Route and Answer mode?
Route shows step 1 only: the full ranked board — every model's predicted odds of getting your question right, and its estimated price. Answer actually sends the question on to the winning model and streams the reply, with its real cost shown.
Why do I have to sign in for Answer mode?
Answering calls a real model through OpenRouter, which costs money. Sign-in is the abuse valve that keeps the demo's budget alive — not a data grab. Route mode never needs an account.
Where are my chats stored?
Signed out: in this browser only — nothing leaves it. Signed in: synced to your account under row-level security, so only you can read them. Export or delete everything at any time in Settings.
Are my chats used for training?
No — not unless you explicitly opt in. The toggle is off by default and lives in Settings; details in the Privacy Policy.
Which model actually answers me?
Whichever of the 17 the router predicts is most likely to get your question right — reaching for a flagship only when the question demands it. The board always shows the pick, its predicted odds, and its price — nothing about the choice is hidden.