Claude Fable 5.2 Cuts Token Prices 20%, but Task Costs May Rise

At a glance

Anthropic's new Claude Fable 5.2 has 20% lower per-token prices than version 5.1. However, the model uses more tokens for reasoning, so the actual bill for a task can go up.

Key points

1. 1

   Input and output token prices are both 20% lower than the previous 5.1 version.

2. 2

   In independent tests, 5.2 used nearly twice as many tokens on the same tasks, raising the cost per task by 56%.

3. 3

   Costs and performance vary a lot depending on the effort setting, which controls how hard the model works.

4. 4

   Anthropic says 5.2 is faster at answering and scores higher on benchmarks.

Why it matters

Developers using Claude Code should look at the actual cost per task, not just the price list.