[bhavsarpratik]

Summary

This adds built-in pricing for Claude Fable 5.2 (claude-fable-5-2) and Claude Mythos 5.2 (claude-mythos-5-2), both released this week.

Rates are per MTok, from the Anthropic pricing page as checked on 2026-11-11:

	Input	5m cache write	1h cache write	Cache hit	Output
Fable 5.2 / Mythos 5.2	$8	$10	$16	$0.20	$40

• Fast mode: both models now use the fast-premium rules, so a fast request is priced at 2× ($16 / $80). Before this change, a fast request on these models had no available cost.
• US inference: both models now use the inference-geo rules, so inference_geo: "us" adds 1.1×, applied on top of fast pricing.
• Tier: both models are in the frontier tier.
• Context window: Fable 5.2 gets a 1M window, as listed on the models overview page. Mythos 5.2's window isn't published, so it stays unknown, the same as earlier Mythos models.
• Review dates: both the pricing and context-window review dates are now 2026-11-11.

Before this change, both models showed no available cost.

Validation

• Full suite (python3 scripts/run-tests): 2671 tests OK, 21 skipped. Run once on each head, b5c9cba and ea4a032.
• py_compile and git diff --check are clean.
• Independent tester probes:
  • Standard, fast, US, and fast+US costs match the published rates for both models and a dated snapshot.
  • Fable 5, Fable 5.1, Mythos 5, and Mythos 5.1 prices are unchanged.
  • The claude-fable-5 rule doesn't catch claude-fable-5-2, and the reverse doesn't happen either.
  • A Mythos 5.2 trace reports context_pct as None, not zero.
• Pricing-semantics review and correctness/regression review are done. Both findings (an unsourced Mythos 5.2 window and stale review dates) were fixed in ea4a032, and a delta review of that commit passed.

🤖 Generated with Claude Code