A pair of posts on X from @lyraxana reports that Anthropic is stealth testing a model called Claude Sonnet 5.5, with the identifier claude-sonnet-5-5. A follow-up post adds reported details about limited partner access, model settings, tool use and pricing. Neither post independently confirms an Anthropic announcement, public release or the authenticity of those specifications.

What the two posts report

The original post says Anthropic is currently “stealth testing” Claude Sonnet 5.5 and gives the identifier claude-sonnet-5-5. It directs readers to the comments for more information.

In the follow-up, @lyraxana says that partners received access for only 24 hours. The report does not identify those partners or explain what the access involved, so it does not establish broad availability or a public test.

Reported model settings and limits

The follow-up attributes several technical details to the reported model:

  • A 1M context window

  • A 128K maximum output

  • Adaptive thinking enabled by default

  • The ability to disable thinking only for low, medium or high settings, but not for xhigh or max

These are details reported by @lyraxana, not specifications confirmed by Anthropic or independently tested in the supplied evidence. The posts also do not explain how the reported thinking modes work in practice.

Reported tool-use change and pricing

The follow-up says that forced tool use is retired. It does not define the change or explain which products, APIs or tools it would affect.

It also lists the following pricing:

  • Input: $2

  • Output: $10

  • Cache reads: $0.20/M

The post does not specify the billing unit for the first two figures or define the notation for cache reads. The figures should therefore be reproduced as reported pricing, not presented as confirmed or final Anthropic pricing.

What “stealth testing” means in this report

“Stealth testing” is the wording @lyraxana uses to describe the reported activity. It may suggest testing that is not being broadly announced or made publicly accessible, but the posts do not explain what the testing involves.

The reported 24-hour partner access could describe a limited evaluation arrangement, but the report does not identify the participants, show how access was provided or establish whether the arrangement is still active. It also does not show that Claude Sonnet 5.5 is ready for release or that the public can access it.

What the report still does not establish

The two posts do not provide independent evidence of:

  • An Anthropic announcement or official position

  • A public release, API listing or general user access

  • Final model specifications or pricing

  • Claude Sonnet 5.5’s capabilities or intended improvements

  • Benchmark results or independent testing

  • The scope and outcome of the reported testing

  • A release date or announcement schedule

A model name, identifier, reported partner access arrangement or listed price does not by itself confirm a completed product launch. The technical details and pricing could be incomplete, provisional, misunderstood or unauthenticated; the supplied evidence does not resolve those possibilities.

How to describe the report accurately

The supplied evidence supports describing this as an unverified report from @lyraxana that Anthropic is testing a model named Claude Sonnet 5.5. The added details remain statements from the post rather than independently confirmed product information.

For now, claude-sonnet-5-5 should be treated as a reported identifier rather than confirmation of a released Anthropic model. An Anthropic announcement, official documentation, an API listing, a model card or reliable independent test results would provide stronger evidence about the model’s status, capabilities, specifications and pricing.

Sources