Claude Opus 5.5 Makes Endless Coding Sessions Cheaper

Anthropic has rolled out Claude Opus 5.5, a new model specifically engineered for the increasingly long and context-heavy coding sessions developers now inflict upon their AI tools. Aggregate usage data shows that context per request has surged 2.6x over the past six months, prompting Anthropic to slash cached token prices by 60% and adjust model behaviour to burn through fewer unnecessary turns on complex tasks.

• Token economics: Input and output token costs are down 20%, while cached token reads dropped by 60%, delivering an estimated 40% overall reduction in running costs for typical workloads.
• Cache preservation: Claude Code updates now prevent routine actions like login refreshes or changing effort levels from breaking your cache, keeping hit rates high despite ballooning context sizes.
• Efficiency gains: Opus 5.5 completes open-ended tasks in fewer turns than its predecessor while generating output over 30% faster, meaning less time spent staring at a progress bar.

Why should I care? Ehhh
Cheaper tokens and better caching are nice, but you still have to read the code your agent writes.

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