What is the best model for code generation?
v5 · 2026-08-244 measured options · top quality 1.000 · from $0.0232/1K requests
Frontier Notes · the reference section
The Measured Answers
Which model should you use for this kind of work, and what does it cost? Below are the live-measured answers — quality, price per 1,000 requests, and latency — re-measured weekly, stated with dates and intervals, from the same measurements that route production traffic. Names that are part of the product are withheld; their numbers are not.
method: how these numbers are made · every page updates when its measurement does
What is the best model for code review?
v4 · 2026-08-246 measured options · top quality 0.972 · from $0.0528/1K requests
What is the best model for data extraction?
v4 · 2026-08-243 measured options · top quality 0.985 · from $0.0238/1K requests
What is the best model for summarization?
v4 · 2026-08-244 measured options · top quality 0.983 · from $0.0345/1K requests
What is the best model for text classification?
v5 · 2026-08-243 measured options · top quality 1.000 · from $0.0042/1K requests
When do you need a reasoning model?
v4 · 2026-08-244 measured options · top quality 1.000 · from $0.0091/1K requests
What is the best model for creative writing?
v4 · 2026-08-243 measured options · top quality 0.900 · from $0.0611/1K requests
What is the best model for rewriting and editing text?
v5 · 2026-08-247 measured options · top quality 0.950 · from $0.3439/1K requests
What is the best model for RAG answers?
v4 · 2026-08-243 measured options · top quality 0.982 · from $0.0058/1K requests
What is the best model for tool calling?
v4 · 2026-08-244 measured options · top quality 0.972 · from $0.1094/1K requests
This section is part of Frontier Notes, Potion Research's publication: the weekly issue covers what moved and what failed; these pages hold the current answers. Every claim links to its method, and negatives are published with the same prominence as wins.