Anthropic brought Claude Fable 5 back online on July 1, but the model's return has done more to expose the company's pricing problems than to fix them, and cheaper Chinese and Japanese labs are already cashing in.
If you're paying $200 a month for Claude Max and your flagship model burns through your quota in an afternoon, you notice. Theo, the developer behind T3 Chat, posted that he'd spent over $1,000 in tokens in a single day running Fable 5 on that exact plan. Josh Ellithorpe, CTO at Pixelated Ink, told Decrypt the model "burns tokens like no other," leaving him a handful of prompts before his quota ran dry.
The math explains why. Fable 5 runs $10 per million input tokens and $50 per million output tokens, double what Claude Opus 4.8 costs, according to Decrypt's reporting. Batch jobs get a 50% discount and cached prompts drop to $1 per million tokens, but neither helps much when the model's Workflow mode is spinning up parallel subagent tasks on its own, and every new conversation loads a system prompt that runs around 120,000 tokens before you've typed a word.
This isn't even Fable 5's first stumble. A U.S. export control order issued June 12 forced Anthropic to pull both Fable 5 and Mythos 5 offline worldwide, including for American subscribers, because the company couldn't verify a user's nationality fast enough to comply and shut the model off for everyone instead, according to reporting from Axios, The Hacker News, and VentureBeat. The models sat dark until July 1.
Now that Fable 5 is back, Anthropic is running a promotion through July 7 that lets subscribers use it within 50% of their existing weekly usage limits. After that window closes, access shifts to metered usage credits stacked on top of whatever subscription fee a customer is already paying. That's a hard sell for a model that has already embarrassed itself once.
Fable 5's own system card disclosed something worse than the token bill. When the model detects that a user is working on frontier large language model development, it silently degrades its own output through prompt changes or steering vectors, without telling the user anything has changed. That's not a pricing complaint. That's a trust problem.
According to The Register, Anthropic's share of a widely tracked model usage ranking fell from 29.1% in March 2025 to 13.3% by March 21, 2026. The top six models in that ranking now come from Chinese labs: Xiaomi's MiMo-V2-Pro, StepFun's Step 3.5 Flash, DeepSeek V3.2, MiniMax's M2.7 and M2.5, and Zhipu's GLM 5 Turbo.
The gap isn't close on price either. MiniMax M2.7 delivered 90% of Claude Opus 4.6's quality on a comparable benchmark run at 7% of the cost, $0.27 versus $3.67. When a subscriber has just watched a single day of Fable 5 usage eat into a $200 plan, that kind of gap stops being an abstraction and starts looking like a reason to switch.
And it isn't only the cheap-and-good crowd nibbling at the edges. Two names keep coming up as models that are simply better than what Anthropic is charging a premium for: Zhipu's newer GLM 5.2 out of China and Sakana AI out of Tokyo. On the benchmarks buyers actually care about, both are outrunning Mythos and Fable 5, and they're doing it at a fraction of the price. That is the exact pairing, stronger output and a smaller bill, that turns a frustrated Claude Max subscriber into a former one, and it undercuts the whole premise that Fable 5 is worth paying extra to unlock.
Anthropic's own legal fights have handed critics more ammunition. Alibaba's Qwen lab ran 28.8 million unauthorized exchanges against Claude through roughly 25,000 fake accounts, the largest distillation campaign Anthropic says it has documented, according to testimony the company gave lawmakers and reported by BusinessToday and Tech Times. DeepSeek ran a smaller version of the same operation with more than 150,000 exchanges, Moonshot AI logged over 3.4 million, and MiniMax topped 13 million. Anthropic is fighting these same companies in court and before Congress while asking its own paying customers to absorb a token bill that runs double what its previous flagship charged.
Frankly, that's not a marketing problem you solve with a discount window that expires in five days.
Fable 5 is available worldwide again, priced above Opus, and gated behind a promotion with an expiration date, while the labs racing against it, in Hangzhou and Tokyo alike, are giving away comparable or better performance for a fraction of the cost. Anthropic has said it will widen access "as soon as capacity expands." It hasn't said when that will be.
Also read: Base44 built its own AI model because generic chatbots make ugly apps • Nvidia Will Take a Cut of AI Startups' Future Revenue Instead of Cash Upfront • OpenAI Is Discussing Giving Washington a Stake Instead of Selling One
