Fable 5.2 Costs 20% Less per Token and About Twice as Much per Task at Max Effort
Fable 5.2's $8 and $40 per-million-token prices are 20% below Fable 5.1's. Artificial Analysis found its average task at maximum effort costs $26.50, about twice its predecessor's.
TL;DR
- Anthropic launched Fable 5.2 on November 9 at $8 per million input tokens and $40 per million output tokens, a fifth below Fable 5.1.
- The cut moves the list price, but the effort setting a buyer picks decides what each task costs.
- Reuters reported that Anthropic's quarterly growth rate has declined sharply and that its IPO roadshow was expected to start November 10.
Anthropic cut the price of its newest model by a fifth and told most buyers to start on an older one.
On November 9, Anthropic launched Fable 5.2 and Mythos 5.2. Mythos is the same underlying model as Fable without certain cybersecurity and biology safeguards, and organizations apply for access to it. Prices per million input and output tokens:
- Fable 5.2 and Mythos 5.2: $8 and $40
- Fable 5.2 fast mode: $16 and $80
- Fable 5.1: $10 and $50
- OpenAI's GPT-6 Astra: $10 and $50
The launch page also says that for most coding and knowledge tasks, "we recommend that users start with Claude Opus 5.5."
That puts Fable 5.2 20% below GPT-6 Astra on list price, Fortune reported, and cached tokens cost $0.20 against OpenAI's $1. Fortune tied the launch to Anthropic's push to re-accelerate revenue growth after a third-quarter slowdown.
The cut looks aimed at consumption revenue, but the bill a buyer pays depends on the effort level the team runs.
Anthropic sells the model on cost per task
Anthropic says Fable 5.2 at High effort matches Opus 5.5's best FrontierCode score for roughly half the cost per task. It also claims output up to 30% faster.
Consumption made up 80% of Anthropic's first-half 2026 revenue, according to Patrick Stähler's analysis of the S-1. Cloud marketplaces carried 48% of revenue in the first half and 44% over nine months.
Stähler found that models released in the prior six months brought in more than 80% of third-quarter revenue. In Stähler's words, "Anthropic meters tokens, while its customers buy work done."
Effort settings decide the bill
Artificial Analysis, an independent benchmarking firm, measured average cost per task at maximum effort:
- Fable 5.2: $26.50
- Fable 5.1: $12.99
- Opus 5.5: $10
The new model costs about twice its predecessor per task, even after the 20% token cut.
The higher cost per task means Fable 5.2 spends more tokens on each task at that setting. Artificial Analysis also found that at lower effort, Fable 5.2 can beat Opus 5.5 at a lower cost per task. The firm ranks the model first on its Intelligence Index.
Two teams paying the same $8 and $40 can see task bills far apart, depending on the effort level they run. The per-task figure was already the number Anthropic and Stacklet sold on, as we covered with the Haiku 5.5 price cut. Anthropic's advice to start most work on Opus 5.5 points the same way: toward the model that, on Artificial Analysis's numbers, costs less per task at maximum effort.
Growth jumped after the last cheaper model
After the cheaper Opus 4.5 shipped, quarter-on-quarter growth rose from 48% to 132%, Stähler found, though the filing credits several factors. On that record, a lower list price may buy volume.
Third-quarter revenue was $17.0 billion against $11.6 billion in the second quarter, per Stähler's reading, with an operating loss of $567 million. That's about 47% growth over the quarter. Reuters reported that Anthropic's quarterly growth rate "has declined sharply."
Fortune, citing The Information, reported that OpenAI's Codex gained ground on price, and OpenAI has been discounting to win Anthropic's top accounts. Reuters reported the IPO roadshow was expected to start November 10.
A rate card can't forecast what Fable 5.2 costs per task
Any buyer that forecasts AI spend from a rate card gets a number that misses the effort level. OpenAI prices GPT-6 Astra by the token too. Anthropic cut list prices with Opus 4.5, and growth jumped the next quarter. It has now cut them again with Fable 5.2, while the launch page argues in cost per task and the invoice still counts tokens.
For a GTM team, the effort level its people run may set the AI bill more than the rate card does. The S-1 names task-based pricing only as a possible future model.
Anthropic meters tokens, while its customers buy work done.
Anthropic's IPO roadshow is expected to start Tuesday, Reuters reports, with quarterly growth that Reuters says has declined sharply.
