MechanismArticle · 43 words
GPT-5.6's new ultra effort level coordinates four agents in parallel by default, trading higher token use for stronger results and faster time-to-result on demanding tasks.
OpenAI · Latent Space
MechanismArticle
Reasoning effort levels are typically implemented one of two ways: applying different length penalties during RLVR keyed to the system-prompt effort label, or fine-tuning the model after RLVR via supervised fine-tuning on effort-labeled target responses.
Raschka lays out the two plausible training mechanisms behind reasoning-effort toggles: RLVR with effort-dependent length penalties, and SFT on effort-labeled data. ✦ AI generated
Sebastian Raschka · Ahead of AI · 2026-07-18 · original ↗
First, we can implement it as part of the RLVR process and apply a different length penalty when different system prompts are used. For example, a high length penalty when "Reasoning effort: low" and a mild or no penalty when "Reasoning effort: high". Second, we can fine-tune the model after RLVR to follow different effort instructions via supervised fine-tuning (SFT).
Read full article ↗excerpt · fair-use quotation
Related moments
MechanismArticle · 216 words
Inkling's reasoning effort is a learned setting between 0 and 1, trained via reinforcement learning so the model associates higher effort with longer chain-of-thought generation, allowing it to match NVIDIA's Nemotron 3 Ultra on Terminal Bench 2.1 while producing roughly a third of the tokens.
Thinking Machines · ByteByteGo Newsletter
AnecdoteArticle · 36 words
Using DeepSeek-V4-Flash-0731's default reasoning level via OpenRouter produced a disappointing pelican image, while raising the reasoning effort to high yielded a much better result.
Author · Simon Willison's Weblog