Blog

A practical comparison for agent workflows, coding, data analysis and computer use, with benchmarks, API prices and a plan for your own evaluation.

Haiku wins on short-request pricing; Flash-Lite is cheaper for long inputs. Current rates, independent benchmarks, and practical choices for automation and business agents.

GPT‑6.1 Sol changes the migration checklist: cheaper cache reads, no none reasoning and Responses for tools. How to test it against Astra and Luna.

OpenAI restored a five-hour window on top of the weekly allowance for Plus users in ChatGPT Work and Codex. For long agents, cost per completed task, checkpoints, and credit options now matter more than a model’s API price.

GPT-5.6 is a routing family, not a universal default. API pricing is now official: promotional Sol is $4/$20, Terra $2/$12, and Luna $0.20/$1.20 per million tokens.