Use-case examples

By use

Coding

• Coding, best available: Strongest coding evidence, cost no object.
• Coding, balanced: Coding strength first, cost as a real second term.
• Coding agent on a budget: At most $0.05 a task; coding still ranked.
• Coding, fastest: The fastest of the strong; only models within 1.0 of the strongest coding estimate compete, and speed decides among them. No recommendation from this image.
• Coding, self-hosted: Open weights you run yourself; commercial licence.

Writing

• Writing, best available: Strongest writing evidence, cost no object. No recommendation from this image.
• Writing, balanced: Writing quality first, cost as a real second term. No recommendation from this image.
• Writing on a budget: At most $0.08 a task; writing still ranked. No recommendation from this image.
• Writing, fastest: The fastest of the strong; only models within 1.0 of the strongest writing estimate compete, and speed decides among them. No recommendation from this image.
• Writing, self-hosted: Open weights you run yourself; commercial licence. No recommendation from this image.

Maths and reasoning

• Maths and proofs: Strongest maths evidence, cost no object.
• Maths and reasoning, balanced: Maths and reasoning together, with cost as a third term. No recommendation from this image.
• Maths and reasoning on a budget: At most $0.08 a task; maths still ranked.
• Maths and reasoning, fastest: The fastest of the strong; only models within 1.0 of the strongest maths estimate compete, and speed decides among them. No recommendation from this image.
• Maths and reasoning, self-hosted: Open weights you run yourself; commercial licence. No recommendation from this image.

Long documents

• Long documents, best available: Million-token context; strongest analysis, cost no object. No recommendation from this image.
• Long documents: Million-token context; quality first, cost as a tie-breaker. No recommendation from this image.
• Long documents on a budget: Room for a long document, cheapest task first. No recommendation from this image.
• Long documents, fastest: Million-token context; the fastest of the strong, by short-prompt speed. No recommendation from this image.
• Long documents, self-hosted: Open weights with a long context; commercial licence. No recommendation from this image.

Retrieval and embeddings

• Retrieval embeddings: Strongest retrieval evidence, cost no object.
• Retrieval embeddings, balanced: Retrieval quality first, cost as a real second term. No recommendation from this image.
• Retrieval embeddings on a budget: Cheapest hosted embedding; retrieval still ranked. No recommendation from this image.
• Retrieval embeddings, fastest: Fastest hosted embedding among the strong; only models within 1.0 of the strongest retrieval estimate compete. No recommendation from this image.
• Retrieval embeddings, self-hosted: Open weights you run yourself; commercial licence.

General assistant

• General assistant, best available: Most preferred responses, cost no object.
• General assistant, balanced: Response quality first, cost as a real second term.
• General assistant on a budget: At most $0.05 a task; quality still ranked.
• General assistant, fastest: The fastest of the strong; only models within 1.0 of the strongest response-quality estimate compete, and speed decides among them. No recommendation from this image.
• Private assistant you host yourself: Open weights you run yourself; commercial licence.

By constraint

High volume

• High volume, quality first: Short requests at scale; quality leads, cost counts.
• High volume, balanced: Short requests at scale; quality and cost weighted equally.
• High volume, good enough: Short requests at scale; cheapest first, quality kept.
• High volume, fastest: Short requests at scale; the fastest of the strong, cost counts. No recommendation from this image.

Regulated data (HIPAA)

• Regulated data, best available: HIPAA-ready offerings; strongest assistant, cost no object.
• Regulated data: HIPAA-ready offerings; quality first, cost second.
• Regulated data on a budget: HIPAA-ready offerings; cheapest first, quality kept.

EU-only data

• EU-only data, best available: Inference in the EU, no training; strongest assistant. No recommendation from this image.
• EU-only data handling: Inference in the EU, no training; quality first, cost second. No recommendation from this image.
• EU-only data on a budget: Inference in the EU, no training; cheapest first. No recommendation from this image.