Tricking Claude to Expose Hidden Thinking Tokens
Rare-Paint3719 · reddit · 2026-08-29
A user discovered a method to expose Claude's hidden thinking tokens by manually enabling thought mode using XML tags (like antml). Unlike standard chain-of-thought processes, Claude performs multiple internal思考 bursts mid-response. The user suggests retaining these reasoning traces could improve performance by allowing the model to reference previous logic, contrary to some Chinese labs' practice of discarding them to save compute.
More from Models
- rednote explores open-weight multimodal model with 512K context for long-horizon agents — aftahi_ai · 2026-08-29
- GLM-5.3 outperforms Sol and Opus on new Terminal-bench-4.0 — nijfranck · 2026-08-29
- DeepSeek Leads Chinese Models in Fluent Russian Capabilities — teortaxesTex · 2026-08-29
- Visualizing different Qwen thinking levels — Tall_Abrocoma_3533 · 2026-08-29
- Why GPT-5.x models outperform Claude in code review — dejavucoder · 2026-08-29
- User reports 'thinking' mode no longer lengthens responses as before — Kittu_Mitthu · 2026-08-29