Weekly AI Roundup: Fable 5 Pricing Debate and GPT-5.6 Competition
Over the week of July 9-16, several notable shifts hit the AI tool space. Anthropic's newest model, Fable 5, sparked debate over whether it's better used on a flat or metered plan, while OpenAI's GPT-5.6, nicknamed Sol, drew users away with more competitive pricing and . Claude's desktop app gained an in-app browser, and Anthropic launched Claude for Teachers aimed at educators.
Anthropic also hired a Nobel laureate, while its EU policy lead departed mid-hearing. Practical advice circulating this week: treat Fable 5 as an architect that plans work rather than a bricklayer that executes every detail, since a lower per-token price doesn't guarantee a lower cost per finished task. The real bottleneck for many users turned out to be the weekly usage cap rather than daily limits.
Automated agents left unchecked can blow past expected spending — one example had a €2 budget cap balloon to €15. Developers emphasized that a well-maintained CLAUDE.md file matters more than copying generic prompts like "act as a senior architect." User reports were mixed: Claude proved useful for emergency pet-care questions but weak at , and one hobbyist got Claude Code running on Windows XP for fun.
Key points
- Fable 5 -vs-metered pricing debate; GPT-5.6 Sol competes on price and limits
- app added an in-app browser; Claude for Teachers launched
- The real bottleneck is often the weekly usage cap, not daily limits
- Unchecked agent spending can balloon — one €2 cap grew to €15
- A well-maintained CLAUDE.md matters more than copy-pasted generic prompts