In general, is it better to use Opus on low reasoning or Sonnet on high reasoning?
retained text
**Edit: For some context, I'm using Claude Models in GitHub Copilot CLI, not Claude Code.** I've been comparing prices, intelligence and token usage between Sonnet 5.5 and Opus 5.5. It appears that Sonnet 5.5 on higher reasoning (namely 'High' and up) is more intelligent than Opus 5.5 on Low reasoning. Opus 5.5 on low is much faster it seems, but slightly less intelligent and a little pricier. I'm unsure if it makes sense to compare models in this way, as there is an element of personality difference that can't be quantified. Does anyone have any advice on what is generally a better idea? (Stats are from artificialintelligence.ai) |Model|Intelligence|Cost per task|Time per Task (Minutes)|Tokens per Task| |:-|:-|:-|:-|:-| |Sonnet 5.5 Medium|41|$0.48|2.1|21k| |Opus 5.5 Low|42|$0.55|1.3|10k| |Sonnet 5.5 High|47|$0.88|3.7|37k| |Opus 5.5 Medium|51|$1.34|3.2|26k| |Sonnet 5.5 xHigh|52|$2.01|7.4|75k| Sonnet xHigh definitely feels like it's not worth the time/cost for the intelligence bump. Opus 5.5 Medium also pretty pricy. Sonnet 5.5 medium -> high is a pretty big cost jump. I'm trying to decide where to settle. If you plot **Intelligence against Task**, **Sonnet 5.5 High** seems the most attractive. If you plot **Intelligence against TIme**, **Opus 5.5 Medium** or **Sonnet 5.5 high** seems most attractive. If you plot **Intelligence against Output tokens**, **Opus 5.5 Medium** seems the most attractive. And then **Opus 5.5 Low** is just **so fast** with such low token usage per task.
document fetched 2026-10-13 00:23 · canonical https://reddit.com/r/ClaudeAI/comments/1x3xpab/in_general_is_it_better_to_use_opus_on_low