Artificial Analysis: Bigger Claude Beats Max Effort
Artificial Analysis finds that, across most of Anthropic's current Claude lineup, running a larger model at lower effort gets more intelligence per dollar than pushing a smaller model to maximum effort.
Benchmarking firm Artificial Analysis reported on X that at nearly every Cost per Task level, the largest available Claude model sits on the performance-versus-cost Pareto frontier for Anthropic's models. For builders, this means raising a smaller model's reasoning effort is usually the less efficient way to get better results.
The firm gave several examples:
- Claude Opus 5.5 at xhigh effort scored 57, ahead of Claude Sonnet 5.5 at max effort at 55. Opus also cost 40% less per task and took about half the time per task.
- Claude Fable 5.2 at high effort beat Opus 5.5 at max effort. The cost and time savings were smaller than in the Opus-versus-Sonnet comparison.
- Fable 5.2 at low effort beat Opus 5.5 at high effort.
The pattern doesn't hold at the cheapest tier. According to Artificial Analysis, every step up from Claude Haiku 5.5 at max effort costs more, so moving up a tier there doesn't save money.
The post doesn't say which benchmark or index produced the scores, and it doesn't give full cost or latency figures for each configuration. It's also unclear how the results would shift for particular workloads or prompt lengths. As a rule of thumb, the finding suggests teams trying to get better results from Claude at a fixed budget should try the next model tier before raising effort settings. The exception is the Haiku tier.