Claude Opus 5.5 vs. Opus 5 on reasoning tasks: Cheaper, faster, but not better

Anthropic recently introduced Claude Opus 5.5, claiming it costs 40% less and operates 30% faster than its predecessor, Opus 5. These claims include Opus 5.5 performing at the level of Claude Fable 5.1, being 40% cheaper for typical workloads, and generating output 30% faster. Additionally, Anthropic reduced the price developers pay to use the model through its API.

To examine these claims, the reporter conducted reasoning task tests on both models. Both Opus 5 and Opus 5.5 received all 28 cells correctly in the logic grid test, which took Opus 5.5 65 seconds and 7,573 output tokens, costing $0.16, versus Opus 5's 108 seconds and 10,621 output tokens, costing $0.27. The logic grid test showed Opus 5.5 to be 43% cheaper and slightly more detailed.

In the constrained orderings test, neither model produced an answer as the model required code execution, which was not permitted. However, Opus 5.5 took 19 minutes and used 112,733 tokens, while Opus 5 took over 25 minutes and used 128,000 tokens. Despite the faster completion time, Opus 5.5's refusal of a response on the 48,000-token limit made the pricing advantage less clear.

The stone game test also showed similar results, with both models correctly answering all three parts. Opus 5.5 was 62% faster and cost 69% less than Opus 5. These findings suggest that Opus 5.5 provides real savings in both execution time and cost for the tested reasoning tasks.