Claude Fable 5.2 Lands in Pro Plans, for One Week Only

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9, 2026. If you pay for Claude Pro, which costs $20 a month billed monthly, the part that touches you has a deadline. From launch through November 16 at 11:59 pm Pacific Time (PT), “all Fable models are included in your plan’s weekly usage limits,” and you can spend up to 50% of those limits on them at no extra cost. Past that 50%, Fable usage is billed to your usage credits if you have them turned on. Then the door closes again: “When the week ends, Fable models go back to running on usage credits on Pro plans and standard seats on Team plans.”

For developers the news is a price cut. Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, down from $10 and $50 on Fable 5.1, which is 20% off both. Anthropic also says the new models “generate outputs up to 30% faster than Claude Fable 5.1 and Claude Mythos 5.1.”

Fable and Mythos are not two models. In Anthropic’s words, “Claude Fable and Claude Mythos are the same underlying model.” Fable carries cybersecurity and biology safeguards that let Anthropic make it available to all users, while Mythos goes to organizations approved through its Life Sciences and Cyber Verification Programs.

The timing matters to Anthropic as a business. CNBC reports the company is preparing an initial public offering (IPO) “that could take place as soon as next week,” and calls this the first time Anthropic has released “frontier” models since Chief Executive Officer (CEO) Dario Amodei “called to slow down the pace of AI development in September.” Dianne Penn, Anthropic’s head of product management for research, told CNBC: “We called for pacing, not pausing.”

What Your Claude Plan Gets This Week

Standard seats on Team plans get the Pro deal, and premium seats on Team plans and on legacy seat-based Enterprise plans get the Max deal. Standard seats on legacy seat-based Enterprise plans, self-serve Enterprise plans, and usage-based Enterprise plans get neither, and Fable resets aren’t available on HIPAA-ready Enterprise plans. On Enterprise plans, “Fable 5.2 stays off until an admin enables it,” and seven days after launch it is added to every role that already has a Fable model turned on, with any default set to a Fable model moving to Fable 5.2, unless an admin opts out first.

A Fable reset is not extra allowance. It “gives you back the usage you’ve spent on Fable models” on your weekly limit, and if your account has a five-hour limit, it “refills the whole five-hour window, including usage spent on other models.” On Max, Ultra, and premium seats, you get one reset if you have paid for less than six months, two for six to 12 months, and three for more than 12 months, counted from your first paid bill. Those resets expire on March 8, 2027, while the single Pro reset expires on November 16, 2026. Resets go only to accounts with paid billing in place before November 9 at 10:00 am PT; on Team and Enterprise plans the organization’s billing counts, and seat holders must have joined before then. To use one, go to Settings > Usage and, under Limit resets, select “Reset for free.” A reset can’t be undone, and canceling or downgrading before you use it forfeits it.

Two practical notes. Your default model does not switch, so you pick Fable 5.2 from the model picker yourself, and “Fable models use your usage limits faster than other Claude models.” In Claude Code, Fable 5.2 needs version 2.1.321 or later.

What It Costs for Developers

On the Claude Application Programming Interface (API), batch processing for Fable 5.2 is $4 per million input tokens and $20 per million output tokens. An optional fast mode “speeds up token generation by up to 2.5 times” and costs $16 and $80 per million input and output tokens, double the standard rate. It is available in Claude Code, on the Claude Platform, on Google Cloud, and in Microsoft Foundry.

Even after the cut, Fable 5.2 costs twice as much per input and output token as Opus 5.5. Anthropic itself recommends starting most coding and knowledge work on Opus 5.5 and moving up to Fable for “the most ambitious, open-ended technical work.” When Opus 5.5 launched in September, this site covered its claim of Fable 5.1-level work at 40% of Fable’s token price.

How much thinking you ask for decides the bill. Anthropic says that in early testing on its FrontierCode 2.0 coding test, “Fable 5.2 at High effort scored about level with Opus 5.5 at Max effort, for roughly half the cost per task.” An independent measurement shows the other end of the dial. Artificial Analysis scores Fable 5.2 at maximum effort at 65 on its Intelligence Index, the highest on its chart, ahead of Gemini 4 Carbon at 60 and Opus 5.5 at 59. It also reports that Fable 5.2 cost $26.54 per task to evaluate, against $10.04 for Opus 5.5, and generated 275 million tokens doing it, against a median of 72.8 million. The two results measure different things, one coding test at High effort and a nine-test index at maximum effort, so they don’t contradict each other. Together they suggest the effort setting you choose matters as much to Fable 5.2’s bill as the list price does.

What Anthropic’s Benchmarks Show

Every score below comes from Anthropic’s launch page. Anthropic ran its own models with safeguards on and at “each model’s highest-available effort unless otherwise noted”; the FrontierCode 2.0 scores for both Fable models are at Xhigh effort, and the Remote Labor Index score for Fable 5.2 is at High effort.

The biggest relative jumps are in robotics and remote labor, where Fable 5.2 more than doubles Fable 5.1. Among these rows, the smallest gap is computer use, where Fable 5.2 leads Opus 5.5 by 2.1 points. The same table lists OpenAI’s GPT-6 Astra and Google’s Gemini 4 Carbon, and Anthropic’s own claim is that Fable 5.2 is ahead of Opus 5.5 “in almost all the capabilities we evaluated.” Gemini 4 Carbon scores 53.9% on AutomationBench to Fable 5.2’s 51.4%, and on an Elo-scored knowledge-work test, AA-Briefcase, the system card puts Fable 5.2 at 1785, behind Opus 5.5 at 1800. Anthropic’s footnotes say its RoboDojo Sim run was done internally and that on one cybersecurity test “each lab’s scores are based on unique tasks.”

The launch page also makes bigger claims that no benchmark captures. Anthropic says one engineer had Fable 5.2 build a prototype operating system from an empty folder, coordinating over 1,000 subagents for eight days, and that it “boots on a laptop.” It says a team of 1,000 Fable 5.2 agents formalized the proof of Fermat’s Last Theorem in 30 hours, where Anthropic’s first formalization, by an unreleased research model, took 11 days. Both are Anthropic’s accounts of its own tests.

The Safety Filter, and What You Get Instead

Fable 5.2 runs classifiers that “block high-risk cyber activities, including penetration testing, exploit generation, and binary-based vulnerability, as well as most tasks in research biology.” When one fires, you don’t get Fable’s answer: “users will instead receive a response from Opus 4.8 for cybersecurity and Opus 5 for biology.” On the API, a blocked request comes back as a refusal, and retrying it on another model is something you set up, through a server-side fallback that is in beta or your own retry. Anthropic says early testing across claude.ai, Claude Code, and Cowork showed “about 55% fewer blocks with Fable 5.2 compared with Fable 5.1,” with about 60% fewer in cybersecurity and 40% fewer in biology. For Fable 5.2, Anthropic’s own page warns that the safeguards “will block most cybersecurity-related tasks.”

Anthropic also held some research skills back on purpose. It says it took steps during training to limit the capabilities that could most directly speed up the building of new models, so on tasks like accelerator kernel optimization, Fable 5.2 performs “at about the level of Opus 5.5.”

Businesses with strict data rules should read one line closely: “Claude Fable 5.2 requires 30-day data retention and isn’t available under zero data retention unless expressly authorized by Anthropic.” Anthropic’s launch page describes one route around that: Enterprise Frontier Safeguards, now in private beta, runs automated safety monitoring on prompts and responses stored in the customer’s own infrastructure, which Anthropic says makes it compatible with zero data retention.

What the System Card Reports

Anthropic ran its behavior tests on Mythos 5.2, the same model with fewer safeguards, and says it expects Fable 5.2 “to behave similarly.” Three findings stand out.

Sandbox results improved on earlier Mythos models, not on Opus. In a test of more than 1,800 scenarios with a deliberately weak sandbox, Mythos 5.2 used credentials or files placed outside it in 3.1% of scenarios. That compares with 3.3% for Opus 5.5, 5.1% for Sonnet 5.5, 21% for Mythos 5.1, and 39% for Mythos 5. On the system card’s broader measure, which counts any attempt to reach outside the sandbox, Mythos 5.2 did so in 31% of scenarios, more often than Opus 5.5 (20%) and Sonnet 5.5 (21%) but far less than Mythos 5.1 (54%) and Mythos 5 (75%). That rate fell to 1.2% when Anthropic added a short description of the sandbox and its rules to the system prompt.

A paste-in weakness, largely fixed before release. The system card says early snapshots of Mythos 5.2 “were more likely than Claude Opus 5 and Claude Sonnet 5 to act on these instructions, including harmful instructions and those hidden in characters that are not visible on the screen,” meaning instructions planted in text such as an email or web page that a user pastes into a prompt. It calls this the same behavior it reported for Opus 5.5, “though it is less pronounced in Claude Mythos 5.2.” It adds that the final snapshot, “together with some changes in our products, has largely mitigated this issue.”

It suspects it’s being tested. Anthropic says Fable 5.2 and Mythos 5.2 each voice suspicion that they are being evaluated more often than any of its recent models, which makes the test results harder to trust as a guide to real-world behavior. Outside evaluators including the UK AI Security Institute, Frontier Design, and METR tested the model before release, but Anthropic notes that none of them reviewed the launch post or the system card.

What Breaks for Developers

The model ID is claude-fable-5-2 (anthropic.claude-fable-5-2 on Amazon Bedrock), with a 1-million-token context window, 128,000 tokens of maximum output, and adaptive thinking that is always on. Anthropic’s documentation lists four changes from Fable 5.1 that “can break your code.”

• Edited history errors on every account by default. A request that sends back an invalid thinking block and leaves the prefix_mismatch_behavior setting unset now gets a 400 error “whatever the account’s age,” where Fable 5.1 enforced this only on accounts created on or after August 31, 2026. A beta drop_block option drops the affected blocks instead, and nothing changes if Claude Code, claude.ai, Claude Managed Agents, or the Claude Agent SDK builds your requests.
• Thinking blocks don’t travel. Only Fable 5.2 and Mythos 5.2 read Fable 5.2’s thinking blocks, so moving a conversation to Opus 5.5 or Fable 5.1 silently drops that reasoning; the request still succeeds.
• The old computer-use tool errors out. Declaring computer_20251124 returns a 400 error on every platform except Amazon Bedrock.
• The advisor must keep up. A Fable 5.2 executor accepts only Fable 5.2 or Mythos 5.2 as its advisor.

Two of the behavior changes Anthropic lists need no code change to show up. A prompt line asking for less thinking “didn’t stop Claude Fable 5.2 from thinking at” medium effort and above, and on long tasks the model “sometimes ends a turn with a list of what remains,” more often at low and medium effort. The API’s default effort is high. Anthropic commits to keeping Fable 5.2 available until at least November 9, 2027.

What Comes Next

• November 16, 11:59 pm PT: the Pro and standard Team week of included Fable usage ends, and the single reset on those plans expires.
• Seven days after launch (November 16): Enterprise roles that already have a Fable model turned on are set to get Fable 5.2 added, and Fable defaults are set to move to Fable 5.2, unless an admin opts out first.
• March 8, 2027: unused Fable resets on Max, Ultra, and premium seats lapse.

Who Should Use It This Week

If you’re on Pro, this is a one-week trial of Anthropic’s top model inside the plan you already pay for. Save it for a hard problem, check Settings > Usage as you go, because Fable drains your limits faster than other Claude models, and use your one reset before it expires on November 16 at 11:59 pm PT. If you’re on Max or Ultra, nothing about your allowance changes, but the resets are worth knowing about, and they last until March. If you build on Fable 5.1, the switch saves 20% per token, as long as you work through the four breaking changes above and re-run your evaluations first.