Anthropic has released documentation and prompting guidelines for its latest frontier model, Claude Opus 5.5. The primary takeaway for businesses and developers building AI automations is speed and efficiency: Claude Opus 5.5 generates output tokens more than 30 percent faster than Claude Opus 5, and it tends to complete identical tasks using fewer tokens.
According to Anthropic, the core control mechanism for the new model is "effort," which dictates how much time the model spends "thinking" before responding. While Opus 5 defaulted to a high effort setting, Opus 5.5 defaults to medium. In Anthropic's testing, Opus 5.5 running at this medium default matches or beats the previous model's high-effort performance across complex coding and knowledge-work evaluations, reducing both latency and cost.
Expanded Capabilities for Knowledge and Visual Work
The release notes detail specific operational improvements that matter for back-office automation. Opus 5.5 is notably stronger at financial modeling tasks, such as finding and fixing errors in valuation workbooks or generating one-page transaction summaries. It is also designed to catch subtle errors in large documents, like a date in a planning thread that falls on the wrong weekday, or a chart in a slide presentation that contradicts the underlying figures.
On the visual front, Anthropic reports a massive leap in efficiency. Opus 5.5 can read values from dense charts at its lowest effort setting more accurately than Opus 5 could at its highest, while using only a small fraction of the output tokens. The model also better understands spatial relationships in images without extra tooling, allowing it to parse flowcharts, spot differences between diagram versions, or accurately read meeting start and end times from calendar screenshots.
For businesses utilizing "computer use"—where an AI operates applications by looking at screenshots over multiple steps—Opus 5.5 at its default medium effort matches the success rate that Opus 5 only achieved at a much higher effort setting. It also sustains long-running autonomous work better, capable of executing multi-hour code audits with parallel subagents.
What This Means for SMB Operations
For small and mid-sized businesses automating their workflows, the transition to Opus 5.5 represents a shift from basic text processing to more autonomous, multi-step operations.
First, the cost and latency improvements change the math on large-scale data processing. Because Opus 5.5 is 30 percent faster and requires fewer tokens to complete a task, businesses can process higher volumes of financial reports, customer support tickets, or inventory data at a lower cost. However, because "thinking" tokens count toward the model's total token limit—which can now be set as high as 128,000 for long, multi-hour autonomous tasks—calibrating the effort dial is critical to avoid paying for unnecessary computation.
Second, the improvements to computer use and visual processing open doors for automating legacy software. Many SMBs rely on industry-specific platforms that lack modern APIs. An AI agent that can reliably operate these applications via screenshots, read calendar interfaces, and parse dense operational charts allows for automations that were previously too brittle to deploy in production.
Finally, the model’s improved communication style makes unattended agents more viable. When running multi-step tasks, Opus 5.5 provides clearer progress updates and final summaries, stating plainly what it did, what it found, and what human intervention it requires.
The rapid evolution of frontier models means that the prompting techniques and system harnesses used just a few months ago quickly become obsolete. Integrating Opus 5.5 isn't just about swapping an API key; it requires recalibrating effort levels, removing outdated "thinking disabled" commands, and adjusting maximum token limits. For businesses relying on AI to run their daily operations, maintaining and upgrading these automated systems is now just as critical as building them in the first place.