Claude Opus 5.5 lowers premium-model pricing and improves output speed, giving enterprise teams a new reason to rerun coding and agent evaluations.
Key takeaways
- Anthropic prices Opus 5.5 at $4 per million input tokens and $20 per million output tokens.
- The company says typical tasks cost 40% less than on Opus 5 and output arrives more than 30% faster.
- Claims about performance and safety still require workload-specific evaluation before production routing changes.
What launched
Anthropic released Claude Opus 5.5 on September 22 as the first model in its 5.5 family. It is available across Claude products and through the API. Anthropic positions the model for coding, knowledge work, computer use and long-running agent tasks, while saying it performs near its larger Fable 5.1 model on much of that work. Independent launch reports confirm the release, pricing and the company’s benchmark framing.
The price cut changes routing math
API pricing is $4 per million input tokens and $20 per million output tokens. Anthropic says a typical task costs 40% less than with Opus 5 because the new model combines lower token rates with greater efficiency. That can matter for coding agents that read large repositories or retry tools repeatedly. Teams should measure cost per accepted task, including cached prompts, tool calls, retries and human review, rather than multiply a headline rate by prompt length.
Speed matters when agents wait
Anthropic claims output is more than 30% faster than Opus 5. Faster generation can reduce idle time inside multi-step workflows, but end-to-end latency also includes retrieval, sandbox startup, external APIs and approvals. A model speedup is most valuable where generation is the bottleneck. Buyers should separate first-token latency, sustained output speed and total job duration, then compare them on identical prompts and tool environments.
Safety claims need operational controls
Anthropic says the model received external testing and safeguards comparable with its most capable systems because of biology and cybersecurity capabilities. That does not replace enterprise controls. Organisations still need least-privilege tools, logging, spending caps, data-retention settings and human approval for consequential actions. Model evaluations should include prompt injection, data leakage, harmful tool use and graceful recovery, not only accuracy benchmarks.
What teams should do now
Re-run production evaluation sets against Opus 5.5, Opus 5 and cheaper models. Record completion quality, latency, total tokens, tool success, reviewer time and escalation rates. The most economical architecture may route complex work to Opus 5.5 while keeping simpler extraction and classification on lower tiers. Claude Opus 5.5 is a credible cost reset, but savings become real only when measured against completed business outcomes.
Facts table
| Release date | 22 September 2026 |
|---|---|
| Input price | $4 per million tokens |
| Output price | $20 per million tokens |
| Claimed task-cost reduction | 40% versus Opus 5 |
| Claimed output-speed gain | More than 30% |
Frequently asked questions
What does Claude Opus 5.5 cost?
$4 per million input tokens and $20 per million output tokens at published API rates.
Is it faster than Opus 5?
Anthropic claims output generation is more than 30% faster.
Is it cheaper for every task?
Not automatically; total cost depends on tokens, retries, tools, caching and review.
Should companies replace existing models immediately?
No. They should compare models on representative tasks and governance requirements first.
Related Lapaas Voice coverage
- NTT DATA expands AI infrastructure operations
- Amazon Leo adds six Ariane 64 launches
- OpenAI model misalignment framework
Verification sources: Anthropic launch TechCrunch MacRumors TechTarget
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
