Anthropic is already using a more capable Claude-family model inside its own operations, but it is not offering that system to outside users. The model is called "Model 2" in the company’s Risk Report from August 2026, where Anthropic describes it as an internal model in the Mythos class.
What Anthropic says Model 2 is
Model 2 is described as an unreleased AI model that performs better overall than every publicly available version of Claude. The report places it in the Mythos class, which means the closest public comparison given in the source is Claude Mythos 5.
Anthropic says Model 2 is slightly stronger overall than Claude Mythos 5, although it is not better in every area. The source makes an important distinction here: this is not described as a major capability leap. Instead, it is presented as a smaller step forward within the same broad class.
On Anthropic’s internal capability index, AECI, Model 2 is about 1.5 points above Mythos 5. The report also says that this improvement is smaller than the jump from Mythos Preview to Mythos 5. That framing matters because it suggests the model is more powerful, but not a dramatic break from the systems Anthropic already discusses publicly.
How the company is using it
Anthropic is relying on Model 2 internally across several kinds of work. The source specifically names coding, data generation, and research and engineering as areas where the company leans on the model heavily.
Some of that use happens through agents that run continuously. The source does not give extra detail about how those agents are configured, what tasks they perform, or how much oversight they receive, so those specifics should not be assumed. What can be said is that Anthropic is not treating Model 2 as a lab-only experiment; it is already part of internal workflows.
The report also says Claude now writes most of the code in Anthropic’s production systems. That statement is separate from the description of Model 2, but it gives context for why an internal model with stronger coding ability would matter. If Claude is already central to production code work inside Anthropic, then a more capable internal model could affect how engineering tasks are carried out within the company.
What the risk review found
Before internal deployment, Model 2 went through an internal review. The source also notes an important limitation: the model was not tested as thoroughly as Mythos 5.
Anthropic says the review did not find new or more worrying misalignments. The company rates the overall risk from misalignment as "low." Those are the risk findings presented in the source, and they are central to understanding why the model is being used inside Anthropic even though it has not been released publicly.
At the same time, the source does not say that Model 2 received the same level of testing as Mythos 5. That leaves a clear distinction between internal deployment and external availability. The model is considered usable within Anthropic’s own environment, but the report does not describe it as ready for public release.
Why the internal-only status matters
There are no plans to release Model 2 externally right now. For customers and developers who use public Claude models, that means Model 2 is not currently part of the available product lineup.
The gap between internal and public models is significant because it shows that Anthropic may operate with capabilities that users cannot access. In this case, the source says the internal model outperforms every publicly available version of Claude, while also stating that the improvement over Mythos 5 is modest rather than a major jump.
That combination creates a narrow but important takeaway. Model 2 appears to be ahead of Anthropic’s public Claude models overall, yet the report does not portray it as a huge new generation. It is stronger, used heavily inside the company, and still being kept behind closed doors.
The main takeaway
Model 2 is best understood as an internal Anthropic system that sits just above Claude Mythos 5 in overall capability, according to the company’s August 2026 Risk Report. It is used for coding, data generation, and research and engineering, including through continuously running agents.
The report says Anthropic found no new or more worrying misalignments and rates the overall misalignment risk as "low." But because Model 2 was not tested as thoroughly as Mythos 5 and has no current external release plan, its role remains internal: a stronger Claude model being used inside Anthropic, not a public product.