New Claude models show strong reasoning, but consume massive thinking tokens
kimmonismus · x · 2026-08-24
Users tested unreleased Claude models (suspected Opus 5.1), noting strong performance on medium reasoning tasks and 3D RL scenarios. A key observation is the heavy consumption of "thinking tokens", frequently hitting the max-tokens limit.
More from Models
- Open source Pelican SVG Env quantifies model drawing abilities — SergioPaniego · 2026-08-24
- Rumor: Kimi K4 to be significantly larger, aiming to rival GPT-6 — iamaliveix · 2026-08-24
- Continuous diffusion language models are making a comeback, Sander Dieleman writes — joao_gante · 2026-08-24
- Testers hint OpenAI is trialing a significantly more capable model — haider1 · 2026-08-24
- Testing Minimax H3's Knowledge of Game and Movie Characters — BoneDaddyMan · 2026-08-24
- Ant Group's Ling-3.0 Open Source: 6 Base Checkpoints Across 3 Training Stages — Affectionate-File-26 · 2026-08-24