Claude Opus 5.5: extra used 3% of the weekly limit

What extra mode showed

The straightforward takeaway from this report is that Claude Opus 5.5 in extra mode may be considerably more economical for ordinary web development than you would expect from a flagship model. The author worked almost continuously throughout the day and used only 3% of the weekly limit.

This was not an agentic marathon. There were no loops, swarms of subagents, or other setups that quickly inflate context and generation: only everyday web application work. In the author's subjective assessment, the quality was no worse than Astra.

Pricing helps explain why the reasoning mode matters here. Anthropic's official Claude pricing documentation listed Opus 5.5 at $4 per million input tokens and $20 per million output tokens at launch. Cache reads cost $0.20 per million tokens, while batch processing cost $2 and $10 respectively.

Output is therefore five times more expensive than input, so lengthy reasoning and verbose generation can damage the economics faster than a large prompt. That aligns well with the second observation: max delivered quality and produced tokens quickly, but generated many of them and took a long time to answer.

Still, 3% of a weekly limit cannot be converted directly into API cost. Product limits, token billing, and caching measure different things. This is one practical episode, not a controlled test with identical tasks and measurements.

Max is no longer the default mode

For routine coding, extra looks like the more rational choice, while max remains reserved for genuinely difficult work. That does not mean max has become worse: additional reasoning pays for itself only when it reduces mistakes, follow-up prompts, or failed edits.

I would first watch response length, the number of repeat passes, and the stability of changes across files. If extra solves a task in one pass, a deeper mode adds delay and usage without a clear return. During a heavy refactor or a long dependency chain, the balance may change.

The signal is not the magical figure of 3%, but a change in habit: maximum reasoning no longer looks like free quality insurance. With Opus 5.5, mode selection becomes part of engineering configuration rather than a slider you always want to push all the way right.

We previously examined how Claude Opus 4.6 balances intelligence, pricing, and configuration options in production. That comparison provides useful context for assessing Opus 5.5’s everyday cost-performance profile.