Anthropic kept the per-token rates for its new model at $2 for every million input tokens and $10 for each million output tokens. This pricing mirrors the previous Sonnet 5 tier and is exactly half of what the competing Opus 5.5 charges. The company also notes that Sonnet 5.5 consumes about 30% fewer tokens per assignment, making the effective cost lower despite the identical headline rates.

Performance on coding benchmarks

In a recent evaluation called Terminal-Bench 4.0, which measures an AI’s ability to complete complex professional tasks autonomously, Sonnet 5.5 achieved a 70.6% success rate. That figure surpasses Opus 5.5’s 66.4% and represents a sizable jump from Sonnet 5’s 10.3% score on the same test. An independent tester, Artificial Analysis, reproduced similar results, reporting 63.6% for Sonnet 5.5 versus 59.6% for Opus 5.5.

Comparisons with rivals

OpenAI recently reduced the price of its GPT-6 Sol model to the same $2/$10 structure, while its mid-tier GPT-5.6 Terra lists at $2 for input and $12 for output. Anthropic did not publish direct benchmarks for Terra. On the GDPval-AA metric, which grades professional work across dozens of occupations, Sonnet 5.5 recorded a score of 1844, essentially tying Opus 5.5’s 1846 and outpacing GPT-6 Sol’s 1487.

Potential trade-offs

When set to the highest effort level, Sonnet 5.5 generates about 193,000 tokens per test case—roughly 60% more than Opus 5.5. This translates to an estimated $7.60 per task, about 50% higher than the cost incurred by Sonnet 5 under the same conditions. Anthropic argues that most users operate at the medium effort default, where the model delivers superior coding results for a fraction of the price of its predecessor. The company also warned that the high-effort figures represent the upper bound of cost, and that typical workloads will see the advertised savings.

Why it matters

Claude Sonnet 5.5 positions Anthropic as a strong contender in the rapidly evolving generative-AI market, offering a blend of speed, cost efficiency, and coding proficiency that challenges both OpenAI and other emerging providers. Its ability to outperform rivals on a professional coding benchmark while maintaining a low price could drive broader adoption in enterprise settings where budget constraints intersect with the need for reliable code generation. The model’s upcoming sibling, Claude Haiku 5.5, aimed at high-volume, cost-sensitive scenarios, suggests Anthropic is expanding its portfolio to cover a wide spectrum of use cases, potentially reshaping pricing dynamics across the industry.