Claude Fable 5.2 on Pro: What to Test Before November 16

The short version: Claude Fable 5.2, Anthropic’s new top model, is included in the $20 Claude Pro plan at no extra cost from today until November 16 at 11:59 pm Pacific, for up to half of your weekly usage. On November 17 it goes back to pay-as-you-go usage credits at $8 per million tokens in and $40 out, which is twice the price of Claude Opus 5.5, the model Anthropic itself tells most people to start with. That gives a small business seven days to find out, on its own work, whether the top model is worth paying for. Here is what to test, and the arithmetic for deciding afterward.

This is genuinely good news, and it is rare. Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, and buried in a help center article is the part that matters to an owner on a single Pro seat: for seven days, Anthropic’s most capable public model is included in the $20 plan, and the clock started running this morning. On November 17 it goes back to pay-as-you-go. Nobody has to buy anything to find out whether it changes their week.

What did Anthropic release today?

Two names, one model. Anthropic says Fable and Mythos are “the same underlying model.” Fable is the version everyone can use, and it carries strong safeguards that block most cybersecurity and laboratory biology work. Mythos is the same model with those guards relaxed, offered only to vetted security and life sciences organizations through verification programs. For a plumbing company, a bookkeeping practice or a dental office, Fable 5.2 is the one that matters.

The headline claims are about difficult technical work. Anthropic says the model formalized the proof of Fermat’s Last Theorem in 30 hours using about 1,000 coordinating agents, designed three working electronic devices from a one-paragraph prompt, and produced output up to 30 percent faster than Fable 5.1. It is also cheaper per token than the model it replaces: Fable 5.1 cost $10 per million input tokens and $50 per million output tokens, and Fable 5.2 costs $8 and $40, a 20 percent cut on both, according to Anthropic’s pricing page, read November 9, 2026. When we covered the Fable 5.1 launch, the headline price did not move at all. This time it did.

The most useful sentence in the announcement is a modest one. Anthropic writes that “for most coding and knowledge tasks, we recommend that users start with Claude Opus 5.5,” and step up to Fable only when the job needs it. That is the company telling you its flagship is a specialist tool. The free week is how you find out whether any of your jobs are specialist jobs.

What does Claude Fable 5.2 cost on the Pro plan?

It depends on the date, and the dates are exact. These terms come from the launch-offer help article and the Fable plan article, both read November 9, 2026.

November 9 through November 16, 11:59 pm Pacific. On a Pro plan ($20 a month, or $17 a month billed annually) and on standard seats of a Team plan, every Fable model is included in your normal weekly usage limits. You can spend up to 50 percent of the week’s allowance on Fable. Accounts with paid billing in place before 10:00 am Pacific on November 9 also get one Fable limit reset, which gives back the Fable usage you have spent. It must be used by November 16 or it lapses.

From November 17. Fable goes back to running on usage credits on Pro and on Team standard seats. Usage credits bill at standard API rates, so Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, charged on top of your subscription. If usage credits are switched off, Fable simply stops being available to you. The help article is explicit that this offer does not include a one-time credit like the one that came with an earlier Fable change in July.

On Max ($100) and Ultra ($200). Fable is a standard part of the plan, up to half of the weekly limit. Those subscribers get one to three extra Fable resets depending on how long they have paid, usable until March 8, 2027.

On Free. No Fable at all, this week or any other.

Two practical traps. First, your default model does not change: you have to pick Fable 5.2 from the model picker, or the week passes without you using it. Second, on Team plans an owner controls which models each person can see, so if your staff are on Team seats, check that Fable 5.2 is switched on for them today, since the week is already running. Anthropic also warns that Fable models burn through usage limits faster than other Claude models, so a long session can eat the 50 percent quicker than you expect. Watch Settings, then Usage.

Is Claude Fable 5.2 better than Opus 5.5 for business work?

On Anthropic’s own table, mostly yes, by a margin that varies a great deal with the kind of work. The three rows closest to what a small business actually does tell different stories, and all three numbers are Anthropic’s, not independent tests.

Finished projects. The Remote Labor Index was built by Scale AI and the Center for AI Safety from 240 real freelance projects, the kind of paid work that gets posted on Upwork, across 23 fields. When it launched in October 2025, the best AI agent completed 2.5 percent of them. Anthropic now reports Fable 5.2 at 41.3 percent at high effort, against 21.3 percent for Opus 5.5 and 17.9 percent for Fable 5.1. We could not confirm that Anthropic scored it the same way the original paper did, so treat the comparison with launch day loosely. The comparison with Opus 5.5 inside the same table is the cleaner one, and it roughly doubles.

Office workflows. AutomationBench tests multistep business processes. Fable 5.2 scored 51.4 percent against 42.5 percent for Opus 5.5. Google’s Gemini 4 Carbon scored 53.9 percent, slightly ahead of Fable, in Anthropic’s own chart. As we found when Mistral claimed a big number on the same test, the office-work test is really a rules test: points are lost to one broken business rule, which is a habit question as much as a model question.

Everyday knowledge work. On the AA-Briefcase rubric, Fable 5.2 scored 66.1 percent to Opus 5.5’s 61.2. On the same benchmark’s head-to-head Elo rating, Opus 5.5 came out slightly ahead, 1800 to 1785. For drafting, summarizing and answering, the cheaper model is already about as good.

Put plainly: the bigger the job, the bigger Fable’s lead. A two-paragraph customer reply will not look different. A day-long project with many moving parts might.

Early testers quoted in the announcement point the same way. Niko Grupen, head of AI research at the legal AI company Harvey, said that in an I-9 employment compliance review, Fable 5.2 “gave the client clearer direction than Fable 5.1,” naming which employees needed attention first and who should fix each problem. I-9 paperwork is something every US employer with a single hire has to keep, which makes it a more useful test than a physics proof.

What should a small business test this week?

The honest way to use a free week is to run the same job through both models and compare the results, not to ask the new one to impress you. Pick work you already understand well enough to judge. Three jobs make a good test, and each should be something Opus or Sonnet has struggled with in the past.

1. A messy numbers job. Export three months of job records or invoices from your accounting software and ask for profit by job type, with the reasoning shown. Anthropic’s testers keep coming back to judgment on spreadsheets: Oracle reported that two other models summed numbers that were 5 to 8 times too high on one query, and Fundamental Research Labs said Fable caught regional figures that did not add up to the reported total. Check the totals yourself against your books. If Fable gets them right where Opus did not, that is real money.

2. A long document you have to act on. A supplier contract, an insurance renewal, a lease. Ask what changed from last year and what you should push back on. Box reported that Fable flagged a loan file where salary deposits did not match the bank statement. You want the same instinct: does it notice the clause that matters, and does it say when it is unsure?

3. A project with several steps. Rebuilding a price list from a pile of supplier sheets, writing a new hire checklist from your old handbook, or turning a quote template into a working spreadsheet. This is where the benchmark gap is widest, so it is where a difference will show if there is one.

For each job, write the same prompt once, run it in Opus 5.5 and in Fable 5.2, and note three things: whether the answer was right, how long it took, and how much of your weekly allowance it used, which you can see in Settings under Usage before and after. That last number tells you what the job would cost once the week ends.

One thing Fable will not do for you: most security checks. Its safeguards block “high-risk cyber activities including penetration testing,” and when a request is blocked, Anthropic answers it with an older model, Opus 4.8, instead. If you ask Fable to probe your own website for weaknesses and the reply seems to come from a different model, that is why. It is not a fault, and it is not something to work around.

Should you pay for Fable after November 16?

Here is the arithmetic, using rates read November 9, 2026. Take a heavy job: Claude reads a document or data set of about 50,000 tokens (very roughly a 70-page contract) and writes about 5,000 tokens of answer and reasoning.

• Fable 5.2 on usage credits: 50,000 tokens at $8 per million is $0.40, plus 5,000 tokens at $40 per million is $0.20. About 60 cents a job.
• Opus 5.5: the same job at $4 and $20 per million is about 30 cents at API rates, and on a Pro plan it costs nothing extra, because Opus is inside your normal limits.
• Max instead of Pro: $100 against $20 is $80 a month more. Divide $80 by 60 cents and you get about 133 heavy jobs a month, roughly six per working day, before Max becomes the cheaper way to run Fable, as long as your Fable use fits inside half of Max’s weekly limit.

Two caveats keep this honest. A long chat costs more than one job, because each new message sends the conversation so far back to the model; starting a fresh chat for each task keeps the bill close to the estimate. And the per-token price is not the whole story. Anthropic says that on its coding test, Fable 5.2 at high effort matched Opus 5.5’s best score “for roughly half the cost per task,” because it used fewer tokens to get there. Fable charges twice as much per token, so it only comes out cheaper on jobs where it needs less than half the tokens. Your test week will tell you whether your work is like that.

For most owners on Pro, the likely answer is this: keep Opus 5.5 as the daily tool, turn on usage credits with a low monthly cap, and send Fable the handful of jobs a month that are worth 60 cents to get right the first time. That mirrors what we found when Opus 5.5 and GPT-6 Sol cut prices: the right model for the job is the one that does it correctly at the lowest cost, not the newest one. If you want a number for what those saved hours are worth, NucoAI, our sister company, offers a free AI efficiency calculator that estimates usable hours and true cost per hour from your AI spend, with no signup needed to see the result.

What the week really buys is information. Most small businesses never get to try a top model on their own paperwork without paying first. For seven days, Anthropic is letting them.

Frequently Asked Questions

Is Claude Fable 5.2 free on the Pro plan?

For one week only. From November 9 through November 16, 2026 at 11:59 pm Pacific, Pro plans and Team standard seats can use up to half of their weekly usage limits on Fable models at no extra cost. From November 17, Fable runs on usage credits billed at API rates, $8 per million input tokens and $40 per million output tokens.

Can I use Claude Fable 5.2 on the free plan?

No. Fable models are only available on paid Claude plans, Pro and above. The free plan runs Sonnet and Haiku, and the launch offer does not include it.

Is Claude Fable 5.2 better than Claude Opus 5.5?

On large, multistep work, Anthropic’s figures show a wide lead, for example 41.3 percent against 21.3 percent on the Remote Labor Index of real freelance projects. On everyday knowledge work the two are close, and Anthropic itself recommends starting with Opus 5.5 for most tasks, since it costs half as much per token.

How much does a Claude Fable 5.2 job cost after the free week?

A heavy job that reads about 50,000 tokens and writes about 5,000 costs roughly 60 cents at the November 9, 2026 rates, against about 30 cents for Opus 5.5. A fresh chat per task keeps the cost close to that, because long conversations resend their history with every message.

If you run a test this week, which job did you give it, and did the top model earn its price?