Instruments and Specifications

Inference as a New Asset Class

An Instrument is a contract between a Consumer and a Supplier.

• When you buy an Instrument, you are buying the right to consume a standardized amount of inference at or above a clear quality and performance floor.

• When a supplier sells an Instrument, they are committing to deliver that inference from any of their models that meet or exceed the Instrument Specifications.

Why this matters

This structure turns inference into a true commodity:

• It replaces provider specific API contracts with grade based contracts.

• It makes Units fungible across suppliers and models, as long as they meet the same spec.

• It enables a single venue where these contracts can be traded with transparent prices.

Trading, Delivery, and Consumption Properties

Each Instrument comes with three categories of properties:

• Trading Properties

  • The atomic Unit size (1 million tokens).

  • How Units can be bought and sold on the CLOB.

  • How Units move between Trading and Consumption via Sweeps.

• Delivery Properties

  • Minimum service level benchmarks for intelligence, latency, throughput, context window, and output length.

  • Requirements around uptime and error rates.

  • Rules for fallback routing and credits if delivery fails.

• Consumption Properties

  • How the 4 hour consumption window works after transfer.

  • How tokens are metered.

  • How rate limits are applied at the API edge.

Deep dive into the below sections to read more about Instruments and their Benchmarks:

• Current Instruments: Chat Prime & Chat Fast - Compares our initial Instruments and helps you choose between speed-optimized Fast Inference and quality-optimized Prime Inference.

• Instrument Specifications - Lists the key metrics - intelligence, latency, throughput, context, output length, and reliability - that define each Instrument’s performance floor, and mentions the current benchmark as of January 2026.

• Evolution of Specifications – Describes how thresholds are periodically updated to track AI progress while keeping suppliers aligned with changing standards.