the priciest Claude model came out at half the price per task
Fable 5.2 costs twice what Opus 5.5 does per token. on FrontierCode 2.0, Fable at high scored 45.3%, Opus at max 45.5%, and Fable did it for roughly half the cost per task
so before you push Opus past high on a hard task, run it on Fable at a lower effort
that's one of Anthropic's own numbers i put into the course below. a few more:
→ 40 turns up to 120K context: ~2.8M tokens sent
→ the same task in 25 turns: ~1.75M
→ 30 turns at 150K: $0.90 in cache reads alone
→ medium effort: 2.5 points under high, ~70% of the cost
→ 2 of 20 problems ate 43% of the spend
honestly the cheapest win in there is one command: /claude-api prompt-audit cut another 9% off Anthropic's support benchmark
every chapter ends with something you paste into Claude Code. start with "where the money goes" ↓
https://www.binance.com/square/post/388046489649540
Fable 5.2 costs twice what Opus 5.5 does per token. on FrontierCode 2.0, Fable at high scored 45.3%, Opus at max 45.5%, and Fable did it for roughly half the cost per task
so before you push Opus past high on a hard task, run it on Fable at a lower effort
that's one of Anthropic's own numbers i put into the course below. a few more:
→ 40 turns up to 120K context: ~2.8M tokens sent
→ the same task in 25 turns: ~1.75M
→ 30 turns at 150K: $0.90 in cache reads alone
→ medium effort: 2.5 points under high, ~70% of the cost
→ 2 of 20 problems ate 43% of the spend
honestly the cheapest win in there is one command: /claude-api prompt-audit cut another 9% off Anthropic's support benchmark
every chapter ends with something you paste into Claude Code. start with "where the money goes" ↓
https://www.binance.com/square/post/388046489649540
