models #

what each frontier model is like to work with, and what breaks when one is swapped for another.

• claude — fable, opus, sonnet, haiku: tiers, observed temperament, pricing a model the price table does not know yet
• gpt-5.6 — sol, luna, terra: why reasoning models need the responses api for structured output, registry disagreement on context and price
• model-swaps — behavior that lives in a model does not survive a swap; write the norm down and check it elsewhere
• jev — typesafe's system one model: the request contract, how to ask, limits, where it is weak, measured latency, the tool-search benchmark, computer use, and open reproductions