What intelligence costs — and how good each tier is.
Two live indices, one page. IQX tracks the market price of each tier, set by sellers competing on the exchange. The Model 500 ranks every model we grade and sets each tier's capability bar. How it's measured →
What each tier is
Tiers are rank bands of the Model 500: Tier 1 the top band, Tier 2 the next, Tier 3 the rest — and each band's floor is that tier's bar. As models are ingested the ranking reconstitutes and the bars track the field.
| # | Model | Tier | Capability | T3 suite | T2 suite | T1 suite |
|---|---|---|---|---|---|---|
| Loading the ranking… | ||||||
Capability = the share of all machine-checkable suite tasks a model passes (0–100). Tiers are rank bands of this list; each band's floor is that tier's bar (≥70% = a suite pass). How grading works →
Every model gets one permanent name
You can't run a market on names like meta-llama/llama-3.3-70b-instruct — every lab spells things differently, renames, and re-versions. So each model in the Model 500 is issued an IXIN (IntellEx Identification Number): a fixed 12-character identifier, the way every security has a CUSIP and every color a Pantone. It reads as foundry · family · generation · scale · alignment, then a registry serial and a check character — so it's both human-readable and unique.
Scale is a coarse band — L = 50–150B here — not a raw parameter count, with U for the closed frontier models that don't disclose it. Capability is never in the name; it's a dated grade.
One stable handle anyone can reference — leaderboards, contracts, research — that never changes when a model is re-graded. Capability lives in a dated fixing, never in the name.
A grade attaches to the IXIN, so it travels across providers. The same model served by three vendors shares one identity; price and speed stay per-listing.
Each IXIN binds to a behavioral fingerprint. If a provider quietly swaps the model behind an API, the fingerprint drifts and we flag it — provenance for intelligence.
Resolve any IXIN over a public API: GET /v1/grades/model/ANTOPS48I01K. Full spec in the methodology.
The index as a product
Live API live
Pull the latest fixings and full history over a simple JSON API — no auth, no key. The same feed this page runs on.
Historical series live
Every hourly fixing since the index began, per grade and composite, with the price source flagged (trade-backed vs quote-backed). Built for backtesting and research.
Benchmark licensing coming soon
Embed the IQX number, settle contracts against it, or license the series as the reference rate for the cost of intelligence — the way markets settle against a published index.
Methodology, governance, and version history: the unit & how prices are set →