What does a 20% price cut buy when the model also invites harder work? Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026 at $8 per 1M input tokens and $40 per 1M output tokens, down from $10 and $50 in the 5.1 generation. In Canadian dollars that is $11.34 CAD per 1M input tokens and $56.72 CAD per 1M output tokens. The held detail is the vendor’s own usage advice. Anthropic still tells most buyers to start with a cheaper model, and that advice is where the saving lives or dies.
The direct answer: Claude Fable 5.2 is generally available at $11.34 CAD per 1M input tokens and $56.72 CAD per 1M output tokens, with an optional fast mode at $22.69 CAD and $113.45 CAD. Mythos 5.2 is the same underlying model, held for vetted cybersecurity and life sciences organisations. A unit price cut of 20% lowers a bill only when the work routed to the model stays the same size.
- What the new price actually covers
- Why cost per task beats price per token
- Who in Canada should pay the Fable rate at all
- The budget rule to set this month
What the new price actually covers
Fable and Mythos are 1 model with 2 access rules. Fable carries stronger safeguards on cybersecurity and biology work, which is why any buyer can use it. Mythos relaxes some of those safeguards for organisations inside Anthropic’s Cyber and Life Sciences Verification Programs. A Canadian research firm that wants Mythos applies first. There is no self-serve path.
The standard rate buys outputs that Anthropic says run up to 30% faster than the 5.1 generation. Fast mode speeds token generation by up to 2.5 times and doubles the rate, to $22.69 CAD per 1M input tokens and $113.45 CAD per 1M output tokens. Speed is a staffing decision priced as a token decision. A team that turns fast mode on for routine work has raised its unit cost by 100% on the day the base rate fell by 20%.
The performance claims are the vendor’s. On Anthropic’s FrontierCode 2.0 test, Fable 5.2 scored 49.1% against 32.4% for Fable 5.1. Treat that as a reason to test the model on your own hard cases. The best AI in the world scored 12% on real professional work shows how far vendor scores sit from finished work in a buyer’s office.
Why cost per task beats price per token
Anthropic’s own efficiency note makes the point better than the rate card does. At High effort, the company says Fable 5.2 scored about level with its cheaper Opus 5.5 at Max effort, for roughly half the cost per task. Read that sentence as a buyer. The comparison that decides your bill is cost per finished task at a set quality, and the effort setting moves that cost more than the generation change does.
Settings are the quiet line in every AI budget this year. Spotify’s plugin cut Claude Code token use by 90% in its own test. Your token bill is a settings problem found a 90% saving inside configuration, with no model change at all. A 20% rate cut is real money. It is the smaller lever on the same bill.
There is a second quiet line. A stronger top model pulls work upward. Teams route the hard migration, the long review and the ambiguous analysis to the flagship because it exists, and each routed task runs longer and reads more context. Gemini 4 Carbon is out. The token price is the small number tracked the same pull on a rival launch this week. The rate fell. The workload grew. The bill is the product of the 2, and only 1 of them made the announcement.
Who in Canada should pay the Fable rate at all
Anthropic answers this in its own launch post. For most coding and knowledge tasks, it recommends starting with Claude Opus 5.5, which it calls its strongest daily driver, and moving up to Fable only when the scope demands it. That is a vendor telling buyers to spend less by default. A Canadian firm should take the advice at face value and build its routing rule on it.
The Fable rate fits the work that fails elsewhere: the migration a mid-tier model cannot finish, the analysis where a wrong answer costs more than the tokens, the long task a person would bill by the week. It does not fit drafting, summarising and routine code help. Those tasks belong on the cheaper tier at the cheaper rate, and a firm that routes them upward is paying flagship prices for commodity work.
The buyer’s wider read sits in the vendor’s own filings. Anthropic lost $80.7B on paper. Its real operating loss was $693M. Buyers should read the second number separates the paper figure from the operating one. A vendor cutting its flagship rate by 20% in IPO season is pricing for volume. Buyers who bring volume discipline to that offer keep the cut. Buyers who bring appetite give it back.
The budget rule to set this month
Set a cost ceiling per task type before the first Fable invoice lands. Pick your 3 most common hard tasks. Run each 1 time on Fable 5.2 at the effort level your team would use in production. Record the full cost of each run in CAD, tokens in, tokens out and any fast mode charge. That recorded figure, and no rate card, is your budget line for the quarter.
The recap holds on 1 figure. The rate is $11.34 CAD per 1M input tokens and $56.72 CAD per 1M output tokens, and the saving is real only for a firm that holds its routing steady while the rate falls.
Set the per-task ceiling this month, while the price cut is new, and read The best AI in the world scored 12% on real professional work for the benchmark check that belongs beside it.
