Claude Fable 5.2 is what Anthropic calls its “most capable model open to all customers”, released on 9 November 2026. In November 2026, it took the top score on the Artificial Analysis Intelligence Index, an independent score built from many tests. At its top setting it scores 65. Gemini 4 Carbon scores 60 and Claude Opus 5.5 scores 59 (Artificial Analysis, Fable 5.2 vs Gemini 4 Carbon, 2026). At max effort, it also costs 2.6 times as much as Opus to run each task.
Fable is the better model. What matters is which jobs are worth the extra cost. Anthropic’s own advice is to start with Opus 5.5 and move to Fable only when Opus falls short. This guide turns that advice into a list of jobs, role by role. Then I show where I use Fable myself, and how I used up a week of it on day one.
Key Takeaways
- Fable 5.2 scores 65 on the Artificial Analysis Intelligence Index, the highest of any model. Opus 5.5 scores 59.
- At max effort it costs US$26.54 per task, against about US$10 for Opus. At high effort it scores 62 for US$9.08.
- Use Fable when a task is long, a mistake is expensive and the result is hard to check.
- On Max, Ultra and premium Team seats, Fable can use up to half of your weekly limit. On Pro and standard Team seats, it runs on paid usage credits. Ration it either way.
Is Claude Fable 5.2 Better Than Opus 5.5? #
Yes on hard work, but not on value for everyday work. In November 2026, Fable 5.2 at max effort scored 65 on version 5.0.1 of the Artificial Analysis Intelligence Index. Opus 5.5 at max scored 59 (Artificial Analysis, 2026).
The jump from the last version was large. Fable 5.1 scored 49 on the same index. Fable 5.2 reached 65 about ten weeks later. It also took the top spot from Gemini 4 Carbon, which had held it for less than two weeks. Carbon scored 59 on the earlier version 5.0 and 60 on version 5.0.1. Google has so far limited Carbon to selected cyber defenders, so most companies cannot use it yet (Google, Gemini 4 Carbon, 2026).
Anthropic is clear about what this means for most people. Its model guide says most work should start with Opus 5.5. You move to Fable 5.2 only “if your evals at xhigh or max effort still fall short on demanding reasoning or long-horizon agentic work” (Anthropic, Choosing the right model, 2026). An eval is simply a test you run on your own tasks to see if the output is good enough.
Anthropic’s launch page shows more benchmark wins for Fable. Read those with care, because a lab chooses which tests to publish. The Artificial Analysis index is independent, so I lead with it.
Citation capsule #
In November 2026, Claude Fable 5.2 at max effort scored 65 on the Artificial Analysis Intelligence Index v5.0.1, the highest of any model. Gemini 4 Carbon scored 60 and Claude Opus 5.5 at max scored 59. Fable 5.1 had scored 49 about ten weeks earlier. Anthropic still advises starting most work on Opus 5.5 and moving to Fable only when Opus falls short. (Artificial Analysis, Anthropic, 2026)
How Much More Does Fable 5.2 Cost Than Opus 5.5? #
About 2.6 times as much per task at max effort, and exactly double per token. Running Fable 5.2 at max costs US$26.54 per Artificial Analysis index task (Artificial Analysis, 2026). Opus 5.5 at max costs US$10.04 on the same index version (Artificial Analysis, Claude Opus 5.5, 2026).
Per token, Fable costs US$8 per million input tokens and US$40 per million output. Opus 5.5 costs US$4 and US$20 (Anthropic, Pricing, 2026). A token is a small chunk of text, about three quarters of a word. Fable is also wordy. It used 275 million tokens to complete the index, against a median of 72.8 million across models (Artificial Analysis, Claude Fable 5.2, 2026).
The useful finding sits in the middle settings. “Effort” is how hard the model thinks before it answers, from low to max. At high effort, Fable 5.2 scores 62 for US$9.08 a task. That beats Opus 5.5 at max on both score and cost. If you bring Fable in, high effort is usually the better buy than max.
Anthropic’s own test results point the same way. On FrontierCode, a coding benchmark Anthropic reports, Fable at high effort came within 0.2 points of Opus at max (45.3% against 45.5%) for about half the cost per task (Anthropic, Introducing Claude Fable 5.2, 2026).
Citation capsule #
Claude Fable 5.2 costs US$8 per million input tokens and US$40 per million output tokens, double Claude Opus 5.5’s US$4 and US$20. On the Artificial Analysis Intelligence Index, Fable 5.2 at max costs US$26.54 per task against US$10.04 for Opus 5.5 at max. At high effort, Fable 5.2 scores 62 for US$9.08, beating Opus 5.5 at max on both score and cost. (Anthropic, Pricing, Artificial Analysis, Artificial Analysis, Claude Opus 5.5, 2026)
Fable 5.2 vs Opus 5.5 at a Glance #
Fable 5.2 is stronger and dearer. Opus 5.5 is the model Anthropic tells you to start with. Anthropic released Fable 5.2 on 9 November 2026, up to 30% faster than Fable 5.1 and 20% cheaper per token (Anthropic, 2026).
| Claude Fable 5.2 | Claude Opus 5.5 | |
|---|---|---|
| Artificial Analysis index (max effort) | 65 | 59 |
| API price per million tokens (input / output) | US$8 / US$40 | US$4 / US$20 |
| Cost per index task (max effort) | US$26.54 | US$10.04 |
| Default effort in the API | High | Medium |
| Anthropic’s best-fit examples | Agent sessions that run for hours, multistep deep research, analysis carried through to a finished document, spreadsheet or deck | Long autonomous coding, large refactoring, complex systems work, computer use |
The best-fit and default-effort rows come from Anthropic’s model guide (Anthropic, Choosing the right model, 2026). In the Claude apps and Claude Code, Fable runs at medium effort by default. It has a 1 million token context window, so it can read very long documents in one go.
Fable 5.2 also has a twin, Mythos 5.2. It is the same model without those cybersecurity and biology safety checks, and only vetted organisations can use it. For how Claude compares with ChatGPT and Gemini for a business in Asia-Pacific, see my ChatGPT vs Claude vs Gemini analysis.
Which Claude Model Should You Use for a Task? #
Use Opus 5.5 by default, and Fable 5.2 only when a task passes three tests. Anthropic’s own guide describes Fable’s best fit as work that runs for hours and ends in a finished document, spreadsheet or deck (Anthropic, 2026).
- Is it long? Many steps, many files, or hours of work.
- Is a mistake expensive? Money, a client, a legal position or a live system is at stake.
- Is the result hard to check? You cannot spot a wrong answer at a glance.
A short email fails all three. Send it to Opus 5.5, or to the cheaper Sonnet 5.5. A loan file review, an acquisition model or a large code change passes all three, so it goes to Fable.
Fable also works well as a reviewer. Run the task on Opus first. Then give the result to Fable and ask it what Opus missed. You pay Fable prices for the review only, not for the whole job.
Anthropic’s docs flag a habit to watch. At low or medium effort, Fable sometimes stops a long task early and hands you a list of what remains, so ask it to finish the whole task. Fable also follows style rules very closely. A hard limit on sentence length can make its writing choppy (Anthropic, What’s new in Claude Fable 5.2, 2026). Relax those rules before you give it your style guide.
Which Roles Get the Most From Fable 5.2? #
The roles where small errors are expensive and hard to spot. In its November 2026 launch, Anthropic quoted early customers in finance, data, law and lending (Anthropic, 2026). Each quote is the customer’s own claim, not an independent test. Together they show a clear pattern.
Finance and FP&A
Spreadsheets that must add up are the clearest case. Jeffrey Liu of Fundamental Research Labs said Fable worked out which metric a sheet meant from its cell formatting. It also caught regional numbers that did not add up to the reported total. It scored 91% on their test, against 86% for the next best model. Routine reports and formatting can stay on Opus.
Data and analytics
A data query has to return the right numbers, and a wrong one can look fine. Liyu Gong, Senior Manager of Machine Learning at Oracle, said two other models wrote SQL that summed country figures wrongly, with results five to eight times too high. Fable’s query returned the expected numbers. For quick queries you can check by eye, Opus is enough.
Legal and compliance
Compliance reviews need a clear list of who must fix what. Niko Grupen, Head of AI Research at Harvey, said that in an employee eligibility (I-9) review, Fable named who needed attention and how to fix each issue. First drafts and plain summaries do not need it.
Lending, operations and document review
Checking documents against each other is slow, careful work. Yashodha Bhavnani, VP of AI Products at Box, said Fable flagged salary deposits in a loan file that did not match the bank statement balance. Pulling fields from one clean document is an Opus job.
Research and investment
In research, the source matters as much as the answer. Ashish Malik, an AI researcher at Balyasny Asset Management, said Fable traced answers to the right sources and did not overstate the evidence. Background reading and first notes are fine on Opus.
Engineering and product
Long builds that start from a loose idea suit Fable. Tom Pritchard, a Staff Engineer at Shopify, said that after a few turns you can tell it “ok, go do that” and expect it to finish. At Uber, Staff Engineer Andrew Mager ran a 10 to 12 hour security audit with about 170 helper agents. Keep Opus as your main coding agent, though. Anthropic still calls Opus 5.5 its “strongest daily driver” for coding (Anthropic, 2026).
Marketing and content
Mostly Opus work, because most marketing tasks are short and easy to check. Fable earns its place on jobs with many strict rules, where every failed attempt costs time and tokens. My example is voiceover scripts, below. For the wider set-up, see how I run Claude Code for marketing.
Citation capsule #
Early Claude Fable 5.2 customers reported gains where small errors are expensive. Oracle said two other models’ SQL summed country figures five to eight times too high, while Fable’s returned the expected results. Fundamental Research Labs said Fable caught regional figures that did not match a reported total, scoring 91% against 86%. Box said it flagged salary deposits that did not match a bank statement. (Anthropic, 2026)
I test each new model on real work and write up only what changes how a company should use it. Join the newsletter to get those notes first.
Where I Use Fable 5.2 Myself #
On work that needs deep analysis and cannot afford a mistake. Everything else stays on Opus 5.5 or Sonnet 5.5.
The jobs I give Fable:
- Deal and investment analysis. When I look at a possible acquisition, a startup investment, my own portfolio or a trade, I use Fable. One missed assumption in those decisions costs real money.
- Voiceover scripts. I generate voiceovers with ElevenLabs, and the scripts carry many rules: where the tone rises and falls, where to , what emotion each line needs. Opus got these wrong and needed several takes. With Fable, I waste far fewer tokens on retries.
- A second pass on app plans. When I plan an application, I write the plan with Opus first. Then I give it to Fable to find the edge cases and gaps I may have missed.
- A check on hard code. When Opus or Sonnet has done a hard coding task and I am not sure it covered every case, Fable reviews it. Hard coding tasks are the other place where Opus fell short for me.
Fable did not help with day-to-day work. It also did not help as the orchestrator, the main agent that plans the work and hands out tasks. I keep Opus in that seat. When a task gets hard, Opus hands it to a Fable subagent, a helper agent that does one job and reports back. That saves a lot of usage. Anthropic describes a similar set-up, where a cheaper model consults Fable only on the hard parts (Anthropic, Choosing the right model, 2026).
Making Fable 5.2 Last the Week #
Use it in short, deep bursts, not as your main agent. On Max, Ultra and premium Team seats, Fable can use up to half of your weekly usage limit. Past that, it runs on usage credits, which you pay for as you go (Anthropic, Claude Fable models on your plan, 2026). I learnt this the hard way. When Fable 5.2 came out, I used it non-stop as my main coding agent. The results were great. By the end of day one of my seven-day cycle, I had used all of my Fable allowance and more than 60% of my usage for the week.
Fable as your main coding agent eats your usage fast. So now I plan the week around it:
- Start and end of the weekly cycle: longer Fable sessions for the big jobs.
- Middle of the week: Fable only as a subagent, for short, deep tasks.
- Every day: Opus as the main agent, Sonnet for the simple work.
Check your usage page as well. At launch, Anthropic gave Max, Ultra and premium seats one to three free Fable resets, based on how long you have subscribed. They lapse on 8 March 2027 (Anthropic, Claude Fable 5.2 launch offers, 2026). Save them for a big job.
If your team is about to learn Claude, plan limits belong in the training. My guide to Claude courses in Singapore covers what a course should teach about choosing between models.
Citation capsule #
On Claude Max, Ultra and premium Team seats, Claude Fable 5.2 can use up to half of the weekly usage limit, and on Pro and standard Team seats it runs on paid usage credits. In Dhawal Shah’s own use, running Fable 5.2 as the main coding agent on its first day used up his whole Fable allowance and more than 60% of the week’s usage. Running Opus 5.5 as the main agent and calling Fable only as a subagent for hard tasks keeps usage under control. (Anthropic, 2026)
What Does Fable 5.2 Cost on Each Claude Plan? #
It depends on the plan. Fable 5.2 is available on Pro, Max, Ultra, Team and Enterprise, but only some plans count it inside their usage limits (Anthropic, Claude Fable models on your plan, 2026).
- Max, Ultra, and premium seats on Team and seat-based Enterprise: Fable is included for up to half of your weekly limit. Past that, it runs on usage credits.
- Pro and standard Team seats: Fable is not part of the plan’s limits. It runs on usage credits, which you pay for as you go. A launch offer included it for up to half of the weekly limit until 16 November 2026 (Anthropic, Claude Fable 5.2 launch offers , 2026).
- Standard seats on seat-based Enterprise: only if your organisation has turned on usage credits.
- Free: not available.
- All plans and the API: Anthropic keeps Fable conversations for 30 days for safety checks. Zero data retention needs Anthropic’s approval (Anthropic, What’s new in Claude Fable 5.2 , 2026).
- Enterprise: Enterprise Frontier Safeguards, in private beta, works with zero data retention (Anthropic , 2026).
- API and usage-based Enterprise: US$8 per million input tokens and US$40 per million output tokens. Batch jobs, which run in the background and return later, cost half that. Fast mode costs US$16 and US$80 and runs up to 2.5 times faster.
For a company, the data retention point matters most. If your contracts promise clients that no AI provider keeps their data, check this with your IT team before anyone pastes client files into Fable. My guide to AI agent governance covers how to set those rules for a team. For what Anthropic’s stock market listing could mean for Claude prices and terms, see my Anthropic IPO analysis.
Frequently Asked Questions #
Is Claude Fable better than Opus? #
On hard work, yes. Fable 5.2 at max effort scores 65 on the Artificial Analysis Intelligence Index, against 59 for Opus 5.5 at max. But it costs about 2.6 times as much per task. Anthropic still recommends Opus 5.5 for most coding and knowledge work, and Fable 5.2 for long, open-ended work where Opus falls short.
Which Claude model should I use? #
Start with Opus 5.5 for most work. Use Sonnet 5.5 or Haiku 5.5 for simple, fast tasks. Move to Fable 5.2 when a task is long, a mistake is expensive and the result is hard to check. A good middle path is to do the work on Opus and ask Fable to review it.
Is Claude Fable 5.2 on the Pro plan? #
Yes, but not inside the plan’s usage limits. On Pro, Fable 5.2 runs on usage credits, which you pay for as you go. A launch offer included it for up to 50% of the weekly limit until 16 November 2026. Max and Ultra plans include Fable for up to 50% of the weekly limit.
What is the difference between Claude Fable 5.2 and Mythos 5.2? #
They are the same underlying model. Fable 5.2 adds safety checks that block most high-risk cybersecurity and most research biology tasks, and it is open to all paid users and API customers. Mythos 5.2 lacks those checks and is limited to organisations vetted through Anthropic’s verification programmes.
Is Claude Fable 5.2 worth the price? #
For a small set of jobs, yes. At high effort it scores 62 on the Artificial Analysis Intelligence Index for US$9.08 a task, which beats Opus 5.5 at max on both score and cost. At max effort it costs US$26.54 a task. Use it for deal analysis, complex reviews and hard code, not everyday questions.
How much does Claude Fable 5.2 cost on the API? #
US$8 per million input tokens and US$40 per million output tokens, down from US$10 and US$50 for Fable 5.1. That is double the price of Opus 5.5, at US$4 and US$20. Batch processing costs half the standard rate, and fast mode costs US$16 and US$80 for up to 2.5 times the speed.
Is Claude Fable 5.2 better than Gemini 4 Carbon? #
On the independent Artificial Analysis Intelligence Index, yes: Fable 5.2 at max effort scores 65 and Gemini 4 Carbon scores 60. Carbon at high effort costs US$7.76 per task, against US$9.08 for Fable at high effort and US$26.54 at max. Google has so far limited Carbon to selected cyber defenders, so most companies cannot use it yet.