OpenRouter recommends Claude Sonnet 5.5 as default

An analysis of Anthropic's Claude 5.5 model tiers—Opus, Sonnet, and Haiku—tested across coding and reasoning benchmarks reveals that Sonnet matches or outperforms Opus at half the price, while Haiku handles routine high-volume tasks economically. The authors advise using Sonnet as a default, Haiku for bulk and sub-agent work, and Opus exclusively for long, ambiguous tasks where mistakes are costly. Open article →