Fable 5 – Median thinking declined in August(twitter.com)
368 points by espeed | 253 comments
An X post by Lon Lundgren with a chart claiming Anthropic's Fable 5 used 22–51% fewer median thinking tokens in August than July on subscription plans.
#llm-performance on Hacker News
2 stories tagged llm-performance, newest first.
GPT-5.5 Codex reasoning-token clustering may be leading to degraded performance(github.com)
339 points by maille | 140 comments
GitHub issue on openai/codex reporting GPT-5.5 reasoning-token counts clustering at 516/1034/1552, possibly degrading performance on complex tasks.