Fable 5 users report 90-minute Max caps and June 23 plan cutoff

• Anthropic said in its launch post that Fable 5 is included on Claude subscription plans only through June 22, while bridgemindai's cutoff post said access will be removed on June 23 until capacity improves.
• Early Max-plan users said Fable 5 can burn through quotas fast: bridgemindai's 90-minute report said a $200 Max plan lasted 90 minutes, while the MTS team burn-rate post described one teammate consuming the equivalent of $1.5k in 10 hours.
• Anthropic reset 5-hour and weekly limits on launch day, according to ClaudeDevs' rate-limit reset, but next-day posts from theo's extra-usage thread and bridgemindai's throughput complaint still described expensive, slow runs.
• Demand looks real even with the complaints: OpenRouter's usage post said Fable was already doing twice Opus 4.8's usage volume, and theo's first-day spend note put one user's first 18 hours at almost $1,000 of inference.
• The launch appears to have pushed some power users toward OpenAI: dylan522p's SemiAnalysis note said frustrated testers gave Codex a serious try, and thsottiaux's Codex consumption post said Codex token usage spiked over the same 48-hour window.

Anthropic's launch post buried the subscription cutoff in its availability section, while the OpenRouter model card spells out the same $10 in, $50 out pricing with a 1M-token context window. You can also check OpenRouter's Anthropic tracker for the early token curve, and the main Hacker News thread for first-day reports from users who hit limits mid-session.

Anthropic launched Fable 5 as the first generally available Mythos-class model, but the included-plan window is temporary. The official announcement says Pro, Max, Team, and seat-based Enterprise subscribers get Fable 5 included only through June 22, after which usage requires credits unless capacity improves.

That detail matters because the public pitch emphasized broad availability first. In the same post, Anthropic says demand will be "very high" and hard to predict, while the OpenRouter model page lists the API price at $10 per million input tokens and $50 per million output tokens.

The early subscription complaints were not about vague "heaviness." They were specific:

• bridgemindai's report said a $200 Max subscription lasted about 90 minutes with Fable 5, and blamed subagents for draining usage.
• theo's first-day spend note said Fable consumed almost $1,000 of inference in the first 18 hours.
• theo's follow-up thread said turning on extra usage burned $56 in about 10 minutes from a single continue on step 12 of a workflow.
• the MTS team post said one teammate hit limits three times in a day and used the equivalent of $1.5k in 10 hours, while half the team hit quota limits on engineering work.
• bridgemindai's launch-day complaint said one user exhausted a $200 Max plan in less than 30 minutes.

Those reports line up with Fable's positioning. Anthropic describes it as a model for long-running autonomous coding and knowledge work in the launch post, and OpenRouter's model card says it is tuned for complex asynchronous tasks with a 1M-token context window. Expensive, multi-step runs are the product.

Anthropic responded fast on launch day. ClaudeDevs' announcement said 5-hour and weekly rate limits had been reset for all users, and theo's reaction post treated the reset like an immediate reprieve.

The next day's complaints were about a different bottleneck. bridgemindai's throughput complaint said Fable was running at 2 tokens per second, and theo's extra-usage thread suggested that even paid overflow usage could disappear quickly once a workflow was already deep into a run.

That leaves two separate pressures in the first 48 hours: hard plan limits, and slow enough throughput that long sessions can feel costly before they finish.

Even with the friction, Fable got immediate volume. OpenRouter's post said Fable was already seeing twice the usage volume of Opus 4.8, with the same daily token usage at twice the price, and linked to OpenRouter's Anthropic tracker.

Some of that demand appears to have rerouted when users hit refusals or limits. dylan522p's SemiAnalysis note said multiple power users tried Mythos and Fable, got angry about refusals, and ended up preferring Codex to Opus 4.8, while thsottiaux's post said Codex token consumption spiked over 48 hours even though OpenAI had not launched anything new.

That makes the first Fable week look less like a clean capacity story and more like a traffic shock across coding tools. petergostev's exchange-rate post framed the subsidy gap bluntly: a $200 Claude plan was still covering usage that would have cost thousands at API rates, at least while the preview window lasts.