Anthropic says: nothing wrong with our usage limits, you're hallucinating
anthropic claude
| Source: HN | Original article
Anthropic has pushed back against a wave of complaints that its Claude model’s token limits are arbitrarily throttling users. In a terse statement posted to its developer forum, the company dismissed the grievances as “hallucinations,” insisting that the weekly and five‑hour rolling caps are functioning as designed and that any perceived tightening is a misreading of the policy.
The backlash began last week when Pro‑plan subscribers reported that sessions on the flagship Opus 4.6 model were ending far sooner than expected, forcing them to switch to the less capable Sonnet 4.6 to stay within their allowance. Users also noted that Anthropic’s public documentation now describes the free‑tier limit as “varying by demand,” a vague qualifier that leaves developers unable to forecast capacity for production workloads.
Why it matters is twofold. First, token limits directly translate into cost and latency for enterprises building on Claude, and unpredictable caps can jeopardise service‑level agreements. Second, the episode highlights a growing tension in the generative‑AI market: providers are balancing resource constraints against the need to keep users on premium tiers, while competitors such as OpenAI and Google offer more transparent quota structures. Anthropic’s defensive tone risks alienating a developer community that already values openness, especially after the recent Claude code leak that revealed internal roadmap details.
What to watch next is Anthropic’s response to the mounting pressure. Analysts expect the company to refine its quota communication, possibly introducing tiered “burst” allowances or clearer peak‑hour pricing. A formal update to the terms of service or a new developer‑portal dashboard could also be on the horizon. Meanwhile, users are likely to test alternative models or negotiate custom contracts, a shift that could reshape Anthropic’s revenue mix and influence broader industry standards for AI usage limits.
Sources
Back to AIPULSEN