Coding sessions are longer and use more context. Claude Opus 5.5 is built with that in mind.

Anthropic estimates that Claude Opus 5.5 costs about 40% less to run than Opus 5 for typical token-billed workloads. This cost reduction specifically targets the shift in developer behavior over the last six months, where coding sessions have become longer and more context-heavy.

Claude Code usage trends

Data from March to September 2026 shows significant changes in how developers use Claude Code:

• Longer execution: Claude works 3.3x longer per session, with 40% more tool calls and 68% fewer interruptions.
• Context growth: Context per request has grown 2.6x, shifting the input-to-output token ratio from 189:1 to 324:1.
• Tool usage: Developers are increasingly using tool servers and skills, requiring more context to be loaded.

Why Opus 5.5 is more cost-efficient

Three factors contribute to the estimated 40% cost reduction:

1. Pricing changes: Input and output tokens are 20% cheaper, while cached token reads are 60% cheaper. Since cache reads constitute a large portion of bills for context-heavy sessions, this reduction is significant.
2. Improved cache utilization: Despite higher context usage, cache misses have decreased by over 50% due to features like persistent cache across effort levels and reduced cache-breaking actions.
3. Fewer turns: Opus 5.5 often completes tasks in fewer turns, especially for complex work. For example, Zeta Labs reported completing twice as many hard tasks at half the cost compared to Opus 5. Additionally, Opus 5.5 generates output 30% faster, reducing idle time.

Best practices for cost optimization

To maximize savings, users should:

• Select their model at the start of a session to avoid cache resets.
• Compact context before stepping away.
• Use /cost to monitor token usage and cache hit rates.
• Ensure context provided is necessary and maximizes cache hits.