Anthropic launched Claude Sonnet 5.5 on September 28 as the company prepares for an IPO. The release reflects a broader shift in how AI technologies are evaluated: while benchmark performance remains important, companies increasingly prioritize cost-effectiveness in delivering results.
Sonnet 5.5 operates faster by over 30% compared to Sonnet 5 and can reduce costs by up to 30% for most tasks despite unchanged token pricing. This cost reduction is achieved through reduced token usage for the same work.
Pricing and Model Positioning
Sonnet 5.5 is priced at $2 per million input tokens and $10 for output tokens, with cache-read tokens at $0.20. Anthropic also offers Opus 5.5, which specializes in complex reasoning tasks, priced at $4 for input tokens and $20 for output tokens. Sonnet is designed for simpler tasks including coding, document generation, and spreadsheets.
This pricing structure reflects Anthropic's focus on enterprise customers, who account for approximately 80% of the company's revenue according to Reuters.
Benchmarks Tell Only Part of the Story
Sonnet 5.5 scores 70.6% on Terminal-Bench 4.0, compared with 10.3% for Sonnet 5. However, raw benchmark scores increasingly fail to capture the full picture. According to the Artificial Analysis leaderboard, Sonnet 5.5 reaches a score of 56 on the intelligence scale at an estimated cost of $7.60 per task, while competing models achieve lower scores at significantly lower costs.
This has prompted companies to adopt mixed strategies, assigning harder tasks to premium models while routing routine work to cheaper systems.
A Decade of Cost Reduction
The shift toward cost-per-task evaluation reflects a longer trend. According to Epoch AI's report from September 22, the costs for achieving a specific level of AI performance have decreased approximately 47% each quarter since 2023, representing roughly a 13-fold yearly reduction.
The Inference Paradox
Lower token costs do not necessarily translate to lower total AI spending. Gartner's August 17 report titled "Gartner Predicts AI Inference Costs Per Agentic Workflow Will Increase More Than Fivefold Through 2028" highlights what researchers call the "inference paradox." Decreased token prices encourage more complex workflows, which increases total token usage and overall expenses.
Global Accessibility Questions
According to Microsoft's Global AI Diffusion Report, low-cost and open-weight models have the potential to increase AI accessibility, particularly in the Global South. However, infrastructure, connectivity, and skills remain determining factors in whether cheaper models will actually be adopted.
What's Next
For enterprise buyers, Sonnet 5.5 demonstrates where the market is heading: the most valuable model is not necessarily the most powerful, but rather the one that performs the required work at a reasonable cost. Haiku 5.5, due in the coming weeks, could accelerate this shift by making high-volume AI tasks cheaper to perform.
