The easiest thing to misread about Fable 5.2 is its bill. Lower base API prices do not put fast mode on the same tariff, or give Pro subscribers unlimited access. Anthropic released Fable 5.2 and Mythos 5.2 on November 9. For users and developers, separating the subscription promotion, API prices and access requirements is more useful than switching every task immediately.
A cheaper token is not necessarily a cheaper finished task
The official price list puts Fable 5.2 at $8 per million input tokens and $40 per million output tokens, versus $10 and $50 for Fable 5.1: a 20% cut in both. Opus 5.5 remains $4 and $20 at base rates, so Fable is not now cheaper than that everyday alternative.
Fable 5.2 fast mode costs $16 per million input tokens and $80 per million output tokens. It is not a free speed switch; check applicable caching and tool charges separately. As a hypothetical base-token calculation, 100,000 input and 10,000 output tokens cost $1.20 in normal mode or $2.40 in fast mode. This excludes other charges and is not a quote for a real task.
Anthropic claims output generation is up to 30% faster than 5.1, with optional fast mode offering up to 2.5 times the generation speed. These are vendor figures, not our measurements or end-to-end delivery guarantees. Our analysis: compare total spend, rework and completion time against the same acceptance criteria, rather than token prices alone.
The Pro promotion has both a deadline and a weekly cap
The official offer lets Pro and Team standard seats spend up to 50% of their weekly usage limits on Fable models through November 16 at 11:59 pm Pacific Time. That is November 17 at 07:59 UTC—not an additional full weekly allowance. Free plans are excluded.
Eligible Pro and Team standard seats also receive one Fable limit reset. Paid billing must have been established before November 9 at 10:00 am Pacific Time, and team members must have joined before then; this reset expires with the promotion. Other plans have different rules. Select Fable in the model picker and check eligibility and consumption under Settings → Usage. After the offer, these seats return to using Fable through usage credits; additional charges depend on your settings.
Fable access does not grant Mythos access
Fable and Mythos 5.2 share an underlying model but differ in safeguards and admission. Mythos is for organizations admitted through relevant cybersecurity or life-sciences verification programs, not automatically for ordinary subscribers. Fable's Claude API ID is claude-fable-5-2. Official specifications list a one-million-token context window, up to 128K output tokens, text-and-image input and text output.
Anthropic lists AWS, Google Cloud and Microsoft Azure access paths, not unconditional availability for every region, account or deployment. Check data policies first: API documentation requires 30-day retention for Covered Models, with zero data retention requiring express Anthropic authorization. Enterprise Frontier Safeguards remains a private enterprise beta with an exception path for eligible customers. It does not make every account zero-retention by default. Obtain organizational approval before sending sensitive material.
Run regression checks, not just a model-name replacement
The migration guide says 5.2 enforces thinking-block prefix checks for every account. Applications that assemble their own messages can get a 400 error after changing earlier system instructions, tools or history and replaying old thinking blocks. Repeating the identical request does not fix that. Users whose history is managed by official products should not treat this as an ordinary chat procedure.
Adaptive thinking stays on, with high effort by default; disabling it is not a way to preserve an old budget. Forced tool calls are also restricted. Start with fixed, sanitized cases covering multi-turn history, tools, errors and cost, then shift a small share of traffic. Our Claude tool page provides a product entry point. The release merits evaluation for demanding technical work, but we have not called the model or tested its performance, and a launch ranking alone is no reason to replace every everyday workflow.