Start with Low Effort, Then Scale Up
mattpocockuk · x · 2026-07-15
The author suggests that when using a model's effort dial, you should start at a low setting and only increase it if quality drops, rather than defaulting to high effort.
The reasoning is that increasing effort essentially means injecting more tokens into the prompt. This might look great on benchmarks—spending 20% more tokens for a 2% gain is good for marketing—but it burns tokens for routine tasks like exploring code repos or modifying tests. This is especially true for more expensive models like Fable.
They also point out that using more tokens increases the risk of attention degradation, which can actually lower output quality. Those "higher effort equals better quality" charts often hide the true costs, latency, and wasted resources.
More from Models
- Kimi K3.1 Pro tops MathArena open models, closing frontier gap to about 6 months — teortaxesTex · 2026-10-26
- Ruff creator Charlie Marsh: frontier models find infinite bugs in any software — charliermarsh · 2026-10-26
- Dev testing one skill across Opus 5.5 and Sol 6.1 finds instruction following 'all over the map' — lucasmeijer · 2026-10-26
- supermemory unveils learner-1, claiming continual learning for any model — julianweisser · 2026-10-26
- Anthropic says Opus 5.5 cuts cost per task by ~40% vs Opus 5 — taufiqintech · 2026-10-26
- AI search APIs ranked: top 12 within 7 points, yet prices differ nearly 7x — FinanceYF5 · 2026-10-26