# It's time to cancel your subscriptions - Anthropic is silently nerfing Claude's reasoning budget while telling you it's the same model — retained record

> Retained reddit record from reddit:ClaudeCode.

Canonical: https://highsignal.app/data/records/190489729792e384

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeCode)  published 2026-09-20 04:17 ingested 2026-09-21 02:38

[original source ↗](https://www.reddit.com/r/ClaudeCode/comments/1wl6uhz/its_time_to_cancel_your_subscriptions_anthropic/)

## retained text



```
Link: https://x.com/Lon/status/2101034933284417614

> A 65-day analysis of 43,000+ Claude Code invocations found that 39% of Fable 5 calls get zero thinking tokens and the median invocation gets just 123 — while benchmarks use 16K-128K. The model's score per thinking token is still climbing at 128K, meaning the capability is there, it's just not being delivered. August saw an 18-50% drop in thinking budget compared to July, with median thinking hitting literal zero for about a week around Aug 22. Anthropic sells "full model access" while quietly dialing down the inference regime behind it, and because the model is non-deterministic, users blame their own prompting instead of the silent nerf. The full breakdown with evidence, methodology, and charts is here:

Frankly I find this offensive as an user - and this is the real thing we should be looking at - not the $/week in usage limits. The actual capability for the limits that we pay for.
```



document fetched 2026-09-21 00:36 · canonical https://reddit.com/r/ClaudeCode/comments/1wl6uhz/its_time_to_cancel_your_subscriptions_anthropic
