Claude Extended Thinking Bug Silently Burns 674k Tokens
forchat4 · reddit · 2026-08-10
A user reported a severe token accounting bug in Claude. Within the same chat session, after executing a second command, the Extended Thinking feature silently consumed massive amounts of tokens in the background, but the UI failed to track it accurately.
- Anomalous Data: The UI showed the response used only 1.1k tokens, but the actual context window usage surged from 70.1k to 744.5k (consuming 674k tokens). The 5-hour limit also jumped from 27% to 84%.
- Impact: Enabling token savings mode did nothing. This bug causes users to rapidly exhaust their usage limits without warning. The author asked the community if others have encountered this and whether disabling Extended Thinking entirely is the only workaround.
More from Models
- Joke: Stop Obsessing Over Massive Models, Many Human Brains Are Just 2B Parameters — yangyi · 2026-08-10
- Alibaba to Release Open-Weight Qwen3.8-Max, Pressuring US Rivals — emmanuelvivier · 2026-08-10
- Nous Research Releases Open-Source Code Model for Local Execution and Agents — emmanuelvivier · 2026-08-10
- Google Reportedly Cancels Gemini 3.5 Pro Development Silently — Left-Hotel904 · 2026-08-10
- Qwen 3.8 Max Faces Backlash: High Benchmark Scores Don't Match Real-World Use — DavidOrzc · 2026-08-10
- Grok Imagine 2.0 Ships, Jumping to #2 Globally in Image Generation — eyishazyer · 2026-08-10