De Vector

Anthropic did release Claude Opus 5.5: correction

Our earlier report that the release was unconfirmed was wrong: Anthropic announced the model itself on 22 September.

Anthropic released Claude Opus 5.5 on 22 September, the first model in a new 5.5 series. In the first version of this article, we wrote that evening that the release had not been confirmed. That was incorrect: the announcement appeared on Anthropic’s website that same day.

Claude Opus 5.5 is a language model from the US AI company Anthropic, the maker of the Claude chatbot. Opus is Claude’s most powerful and expensive line, intended for complex work such as programming, analysis and tasks in which the model takes steps independently on a computer. Opus 5.5 follows Opus 5. According to Anthropic, it performs at the level of the larger Claude Fable 5.1 model on most tasks.

Anthropic gives a price of 4 dollars per million input tokens (pieces of text) and 20 dollars per million output tokens. For typical use, the model is 40 per cent cheaper than Opus 5, according to the company, and writes its answers more than 30 per cent faster. The model is available through Anthropic’s own platform and the cloud services of Amazon, Google and Microsoft, under the name claude-opus-5-5.

What remained unconfirmed: our first version concerned whether Anthropic had already released Opus 5.5 publicly, and whether the model was cheaper than its predecessor. At that point, we found no announcement in the Anthropic pages we consulted, and the details came mainly from posts on X and an analysis by a software company. Both claims turned out to be correct. Our editorial team had seen the wrong or an outdated version of the sources and should have reported that as uncertainty, not as a denial.

Consulted for this revised version: Anthropic’s announcement and product page, the Opus 5.5 system card (the technical report containing test results), Anthropic’s status page and reporting from, among others, TechCrunch. For the first version, we used Anthropic’s product page, release notes and status page, as well as a blog by Valletta Software.

Why this matters: models such as Opus 5.5 underpin much of the software used by Dutch companies and developers, from programming assistants to customer service. A lower price and higher speed make such applications cheaper. Anthropic also says that Opus 5.5 scores better in its own safety tests and is the first Opus model with additional safeguards against misuse in cyber security and biology.

What remains unknown: the performance and safety figures come mainly from tests by Anthropic itself. The company itself writes that differences in such rankings are becoming a less reliable measure. Independent comparisons still need to establish how significant the improvement is in practice. The smaller Sonnet 5.5 appeared on 28 September; Haiku 5.5 has been announced for the coming weeks, without a date.

Respond to this article

← Back to the edition