The September 22 announcement
Anthropic launched Claude Opus 5.5 as the first member of its 5.5 family. The company describes stronger coding and professional work, clearer communication, and more efficient extended tasks. It estimates a 40% cost reduction against Opus 5 for typical token-billed workloads at default settings. That is a vendor estimate, not a guarantee for each user's bill.
The announcement also says output generation is more than 30% faster than Opus 5 and that five-hour limits are increasing on specified subscription plans. Sonnet 5.5 and Haiku 5.5 were described as forthcoming; the launch post did not make them available that day.
Separate task cost from token price
Anthropic's Opus product page lists standard pricing of $4 per million input tokens, $20 per million output tokens, and $0.20 per million cache-read tokens. The input and output rates are 20% below Opus 5; cache reads are 60% cheaper. These rates help explain why its estimate for a whole task differs from the reduction in the standard token rates.
Our advice is to compare invoices or recorded API usage for the same completed workload. The proportions of fresh input, cached context, generated output, and retries all matter. A reduction in one category does not automatically translate into the same reduction in the total. Subscription limits should be checked separately from API prices.
Why longer jobs need a practical trial
A code migration or a substantial report has many opportunities to save time, but also many opportunities to leave something unfinished. For a useful pilot, select one task with known acceptance conditions and retain the starting files. Measure whether the final artifact passes review, how much correction it needs, and how long a human spends supervising it.
A faster stream of text can be helpful, but it does not directly measure the time required to obtain a working result. Customer examples in a launch post can suggest tasks to test; they do not establish the outcome on your repository or document collection. VSike has not run an independent Opus 5.5 benchmark.
Check access and routing as well as capability
The launch identifies claude-opus-5-5 as the API model name and describes safeguards that can route some sensitive categories to other models. For a technical evaluation, preserve any routing information the product exposes and review the applicable integration guidance. A selected model name alone may not explain every result in those categories.
Our suggested adoption decision is task-specific: try the release on a representative job, inspect the result, and compare actual resource use. For a direct API comparison, read GPT-6.1 Sol versus Opus 5.5. The evaluation method from our earlier series can also be adapted to a Claude comparison.