--- title: How Anthropic Fumbled the Fable 5 Launch description: Fable 5 was pulled by a US export order three days after launch. Now GPT-5.6 has closed Anthropic's window to sell it at API pricing. author: Darie Nani (Editor-in-Chief) date: 2026-07-14T11:50:17.991Z updated: 2026-07-14T11:50:18.002Z canonical: https://www.sovereignmagazine.com/article/anthropic-fable-5-fumbled-launch image: https://cdn.nanimediahouse.com/fable-cartoon-3.webp categories: Artificial Intelligence, Politics content_type: Analysis region: United States publication: Sovereign Magazine about: - type: Organization name: Anthropic --- On Sunday Anthropic extended free access to Claude Fable 5 until July 19, the third deadline it has set for the model in two weeks. The plan to take Fable off subscriptions and sell it as a metered, pay-per-token product appears finished, though Anthropic has yet to say what happens after July 19. That plan dates from the model's July 1 relaunch: Fable 5 was to stay in paid plans until July 7, then come off subscriptions and move onto prepaid usage credits billed at API rates of $10 per million input tokens and $50 per million output. Those rates are double Opus 4.8's $5 and $25 and the highest on Anthropic's price list. The window for that plan closed in a sequence largely of Anthropic's own making. The company spent the spring publicizing how dangerous its new model was, and three days after launch the US government answered with export controls that forced Fable offline for 18 days. OpenAI released GPT-5.6 into that gap, priced far below Fable, and Anthropic has been extending free access a week at a time ever since. ## When Hyping Up Your Next Model Goes Very Sideways From April, executives described the underlying system, codenamed Mythos, as too dangerous for general release, and in May chief executive Dario Amodei warned of a "moment of danger." On June 9 the company shipped two versions: Fable 5, with added safety classifiers, for the public, and a restricted Mythos 5 for vetted organizations. Everything it had [said about Mythos](https://www.sovereignmagazine.com/article/claude-mythos-anthropic-new-model) became a liability within three days. On June 12 at 5:21pm ET, the Commerce Department's Bureau of Industry and Security ordered an export license for any foreign national to access either model. The trigger was a jailbreak: Amazon researchers had demonstrated a prompt sequence that produced information useful in a cyberattack, and Andy Jassy had raised it with the White House. Anthropic could not verify user nationality in real time, so it disabled both models worldwide itself, saying it "must [abruptly disable Fable 5 and Mythos 5](https://www.anthropic.com/news/fable-mythos-access) for all our customers to ensure compliance." It disputed the order, calling the finding "a narrow potential jailbreak" and not cause to recall a model "deployed to hundreds of millions of people." TechCrunch's reading at the time was that Anthropic's own safety campaign had backfired, handing the government the pretext to pull its most powerful model. The model sat dark for 18 days before Commerce Secretary Howard Lutnick withdrew the controls on June 30. Fable [returned worldwide on July 1](https://www.sovereignmagazine.com/article/anthropic-fable-5-returns-export-ban-lifted) with new cybersecurity classifiers, and some routine coding work falls back to Opus 4.8 while their false positives are tuned. ## Claude Fable 5 API Pricing vs GPT-5.6 OpenAI had trailed Anthropic for more than a year and moved while Fable was offline: it ran a government-coordinated preview of GPT-5.6 from June 26 with around 20 vetted organizations, then released the family on July 9, two days after Fable's original subscription cutoff. GPT-5.6 comes in three tiers: Luna, Terra and Sol. On OpenAI's own numbers, Sol scores 88.8% on Terminal-Bench 2.1 against Opus 4.8's 78.9%, and Artificial Analysis's aggregated Intelligence Index puts it at 59 to Fable 5's 60 at roughly a third of the cost per task. OpenAI says even Luna, the budget tier, outperforms Opus 4.8. No independent head-to-head test exists, so these are OpenAI's claims and one aggregated index. Sol's API price is $5 and $30 per million tokens, against Fable's $10 and $50. The usage terms favor OpenAI too. Its Codex coding tool meters a quota pool separate from ChatGPT, while Claude Code and Claude chat draw from one shared weekly pool, a design developers call the "chat tax." Sam Altman has tied Codex limit resets to user-growth milestones, and OpenAI has offered enterprise Claude Code customers two months of free Codex to switch. ## Claude Usage Limits and the Moving Deadline The July 1 relaunch came with a promotion: paid subscribers, from Pro at $17 to $20 a month up to the Max plans at $100 and $200, got Fable 5 [free up to half their weekly usage limits](https://www.anthropic.com/news/redeploying-fable-5) through July 7. Anthropic never called it compensation; users and the press did. Anthropic moved that deadline hours before it arrived, extending free access to July 12. Five days later it [pushed the date to July 19](https://x.com/claudeai/status/2076351399999557669), with the deadline it had just set due to expire that night. Fable stays on the API at $10 and $50 throughout. The extensions have angered the customers Anthropic is trying to keep. Subscribers plan work around each deadline and burn their weekly usage to get value from Fable before it disappears, and then the deadline moves. Two extensions in five days train users to disbelieve the next date while punishing anyone who took it seriously. On Hacker News, one developer wrote, "I'd been [working non-stop all weekend](https://news.ycombinator.com/item?id=48882730) to use Fable before it disappeared tomorrow." Another asked Anthropic to "decide on what their plan is with Fable." A third called the company "Deeply unserious... flip flopping on policy every week," and others described pulling Fable from subscriptions after offering it as a "bait and switch." One wrote: "Just cancel and switch to OpenAI." A separate thread read the July 19 date as ["more like an operational checkpoint than a real sunset date."](https://news.ycombinator.com/item?id=48884990) Buyers are making the same calculation. Writing in Forbes on June 29, analyst Ron Schmelzer argued that companies may question whether they can trust a model that was made unavailable so quickly, and that Anthropic faces a harder sales conversation having lost market momentum. The friction has reached enterprise directly: Microsoft removed Fable 5 from Copilot's model picker on June 10 over a mandatory 30-day data-retention rule that conflicts with its zero-retention policy. Capacity is Anthropic's recurring explanation. In April it cut OpenClaw and other [third-party agent frameworks off flat-rate plans](https://www.sovereignmagazine.com/article/anthropic-blocks-openclaw-claude-subscriptions), with Boris Cherny citing "outsized strain" on its infrastructure. On Fable, Thariq Shihipar of the Claude Code team [has said](https://x.com/trq212/status/2072814903170408784) the model "will come off subscriptions after July 7th" but that Anthropic aims "to restore Fable as a standard part of our subscriptions as soon as capacity allows." A company spokesperson gave Wired the same formula a week later: Fable will return "when sufficient capacity allows," and "as quickly as we can." No date is attached to either. That leaves Anthropic the choice it keeps putting off: carry the cost of Fable inside its subscriptions, or watch subscribers leave for a model that costs half as much and is close enough on quality. ## FAQ **Q: Is Claude Fable 5 coming back to subscriptions?** Anthropic has not committed to a date. Thariq Shihipar of the Claude Code team says the company aims to restore Fable "as a standard part of our subscriptions as soon as capacity allows." The model was due to come off paid plans after July 7, but free access has been extended twice and now runs to July 19. The metered API-credit plan announced on July 1 has yet to take effect. **Q: Why was Claude Fable 5 taken down?** On June 12 the Commerce Department ordered an export license for any foreign national to access Fable 5 or Mythos 5, after Amazon researchers showed a jailbreak that produced cyberattack-useful output. Anthropic could not check nationality in real time, so it disabled both models worldwide for 18 days. The controls were withdrawn on June 30 and Fable returned on July 1. **Q: Is Claude Fable 5 free right now?** For paid subscribers, yes, but only until the current deadline. Anthropic has offered Fable 5 free up to half of weekly usage limits since July 1, and has moved the end date twice, from July 7 to July 12 to July 19. On the API, Fable is never free: it bills at $10 per million input tokens and $50 per million output. **Q: How much does the Claude Fable 5 API cost compared with GPT-5.6?** Fable 5 costs $10 per million input tokens and $50 per million output, the priciest model Anthropic lists and double Opus 4.8. OpenAI's top GPT-5.6 tier, Sol, is $5 and $30, roughly a third of the cost per task on Artificial Analysis's index while scoring within a point of Fable on that index.