Claude Opus 5.5, Anthropic’s new flagship model, has been available since 22 September 2026. The publisher promises a model that is faster, more economical and cheaper than Opus 5. Six days later, Claude Sonnet 5.5 completed the range.
Both launches directly concern SMEs that automate customer support, office tasks or development. Here are the verified facts and their practical consequences.
Key points
- Anthropic launched Claude Opus 5.5 on 22 September 2026, then Claude Sonnet 5.5 on 28 September 2026.
- Opus 5.5 costs 4 dollars per million input tokens and 20 dollars per million output tokens, 20% less than Opus 5.
- Anthropic also announces throughput more than 30% higher and a typical usage cost reduced by about 40%.
- Sonnet 5.5 costs 2 dollars for input and 10 dollars for output. Anthropic then plans Claude Haiku 5.5.
Claude Opus 5.5: what Anthropic announced
Opus 5.5 opens the 5.5 family. According to Anthropic, it reaches, on most tasks, the level of Claude Fable 5.1, its most advanced model. MacRumors and BNN Bloomberg relayed this comparison on the day of the launch.
Businesses access it through the Claude platform, but also through AWS, Google Cloud and Microsoft Azure. A company already a customer of one of these clouds therefore keeps its provider and its billing.
Anthropic also stresses the writing style. The model puts the important information at the start of its answers and reduces jargon. This detail matters for customer letters, summaries and content.
Price, speed, economy: Claude Opus 5.5 figures
The announced reduction reaches 20% on the token price. Opus 5 cost 5 dollars for input and 25 dollars for output, per million tokens. Opus 5.5 now costs 4 and 20 dollars.
The real saving, according to the publisher, goes beyond this percentage. Opus 5.5, it says, completes a task with fewer steps and fewer tokens. Anthropic puts the gain at about 40% on typical usage. This figure comes from the publisher: check it against your own tasks.
- Opus 5.5: 4 dollars for input and 20 dollars for output, per million tokens.
- Sonnet 5.5: 2 dollars for input and 10 dollars for output, per million tokens.
- Speed: each model generates text more than 30% faster than its predecessor, according to Anthropic.
Sonnet 5.5 follows the same economy logic. SiliconANGLE reports a cost-per-task reduction of about 30% in typical usage, because the model consumes fewer tokens.
Claude Opus 5.5 or Sonnet 5.5: which model for which use?
Sonnet 5.5 plays the role of the everyday model. It handles support, writing, document analysis and a large part of development. According to SiliconANGLE, Zendesk processed its tickets 20% faster during its tests, with fewer wrong decisions.
Claude Opus 5.5, by contrast, keeps the advantage on long, open-ended work that requires sustained judgement: code migration, financial analysis, coordination of several agents. Finally, Anthropic intends Haiku 5.5 for high-volume applications on a tight budget.
The good practice is therefore to split tasks. The economical model handles the routine, while the premium model intervenes on difficult cases. This split reduces the bill without lowering quality.
What this changes for an SME
First effect: the budget. At equal quality, the same envelope funds more automated tasks. Second effect: speed. An assistant that answers faster fits better into a customer journey or an internal tool.
Third effect: competition works in your favour. OpenAI launched GPT-6 Sol and GPT-6 Luna the same day, with lower prices. But each publisher publishes tests that favour its own model. Only your own use cases settle which model is best.
Last effect: safety becomes a selection criterion. An AI agent acts inside your tools, and its caution matters as much as its intelligence. Anthropic states that external evaluators, including the METR organisation, tested Opus 5.5 before its release.
The publisher also announces a reduction of about 85% in attempts to bypass limits compared with Opus 5. BNN Bloomberg repeats this figure. It comes, however, from Anthropic’s own tests and does not replace your own safeguards: limited rights, human validation, an action log.
- First replay your current scenarios on Sonnet 5.5, before considering Claude Opus 5.5.
- Compare the cost per completed task rather than the price per token.
- Plan a technical layer that can switch models without rebuilding the application.
- Finally, keep human validation on actions that commit the company.
AISYSNEXT’s view
At AISYSNEXT, we recommend not tying a project to a single publisher. Models change quickly: Anthropic and OpenAI each renewed their range in September 2026. Your data, your business rules and your tests remain. Build your project around these three assets.
Our artificial intelligence and BI integration team helps choose the right model for each task. Our custom development team then connects this model to your applications. For content, the same rule applies: AI assists, then a human reviews. Our digital marketing team supports you on this point.
See all our services to place AI within a broader digital project.
Frequently asked questions
Is Claude Opus 5.5 better than GPT-6?
No universal answer exists. Anthropic and OpenAI each publish comparisons that favour their own model. Test both on your documents and processes, then compare quality and cost per task.
Where can you find Claude 5.5?
Anthropic distributes Claude Opus 5.5 and Sonnet 5.5 on its own platform, as well as on AWS, Google Cloud and Microsoft Azure.
When is Claude Haiku 5.5 released?
On 28 September 2026, Anthropic announced Haiku 5.5 for the following weeks, without giving a precise date. The model targets high-volume applications.
Torn between several models for an assistant or AI agent project? Contact our expert team: we will then frame a comparative test with you on your real cases.
