Verifiable routing intelligence for the same model across every provider endpoint — mined, replayed, and verified.
We don't rank models. We rank provider endpoints for the same model.
A decentralized routing-intelligence market — not a dashboard.
Developers access models through multi-provider gateways. The same model name can perform very differently depending on which endpoint serves it.
Model selection became a routing decision — still made on stale benchmarks, anecdotes, and brittle heuristics.
The output isn't a leaderboard. It's production-usable routing guidance the same model can be served by.
The best provider endpoint to send traffic to right now.
A ranked backup sequence for graceful degradation.
A production-ready policy that drops into the gateway.
Per cell, miners submit telemetry, a routing signal, and a proof bundle (prompt/response hashes, timing, schema checks).
→Validators replay sampled tasks under pinned params. Out-of-tolerance trips the Gate to 0 — no LLM judge.
→Emissions flow proportionally to the most accurate, fresh, and useful routing signals.
cell = model × provider-endpoint × task-typeOut-of-tolerance replay invalidates the whole submission. Gate ∈ {0,1}.
Coverage is multiplied by accuracy — scale without truth earns nothing.
Rewards are proportional, not winner-takes-all, with a ~15% weight cap.
Provider quality changes constantly, so routing intelligence must be refreshed continuously.
Validators can replay challenge tasks and verify latency, success, schema, and routing quality.
Better routing lowers cost and latency and raises reliability in production.
Many miners compete to generate better signals while validators continuously verify them.
Not a leaderboard subscription — a shared intelligence layer that replaces fragmented private benchmark stacks.
Recruited from LLM infra, eval/benchmark, and competitive builder communities. Early emission upside on high-value cells.
Technically credible operators who run replay infra. Early position, staking economics, governance influence.
Agent builders & inference teams. Free/discounted early access in exchange for the demand that makes signals worth producing.
Miners compete → validators make signals trustworthy → consumers create the demand. That loop is the go-to-market.
We rank provider endpoints for the same model — then turn that into routing decisions.
ProviderRank · A Bittensor Subnet for Verifiable Provider Routing Intelligence