Cache-read multiplier is wrong on Claude Fable 5.1 / Mythos 5.1 (0.025x, not 0.1x) #74

Problem

The per-turn economics line (line 2) uses a flat cache-read multiplier of 0.1x for every Claude model (compute_turn_breakdown in tokenline.sh).

Anthropic's pricing page now bills cache hits and refreshes on Claude Fable 5.1 and Claude Mythos 5.1 at 0.025x base input ($0.25 / MTok against $10 / MTok base). Every other model keeps the standard 0.1x.

On those two models tokenline overstates eq and understates saving %. Example turn (100k read, 5k write, 1k new, 2k output):

Still correct (checked against the pricing table)

• 5m cache write 1.25x, 1h cache write 2x: unchanged for every model.
• Output 5x: output/input is 5 for every model on the table (Fable 50/10, Opus 25/5, Sonnet 10/2, Haiku 5/1).

Fix

Make read_mult model-aware: read model.id in the single jq pass and select 0.025x when the id (or display name) is Fable 5.1 / Mythos 5.1.

Rust port

The same rule has to land in the Rust parity port (#27): economics, input parsing (model.id) and the golden fixtures.