Anthropic Has a More Capable AI Model, But You Can’t Use It Yet

Anthropic has revealed that it is already using an unreleased AI model called Model 2 internally, and the company says it performs better than its current Mythos 5 model on several internal tasks. However, there are currently no plans to make Model 2 available to the public.

Model 2 Scores Higher Than Mythos 5

The biggest performance difference appears in Anthropic’s internal CoBench v2 benchmark, which contains 449 real research and engineering problems previously solved by Anthropic employees.

Model 2 scored 62.8%, compared with 50.3% for Mythos 5 and 54.8% for Mythos Preview.

For comparison, Claude Opus 4.7 scored 27.4%, while Opus 4.6 scored 15.6% on the same test. Anthropic says a model capable of fully replacing its technical research staff would need to reach around 85%, meaning Model 2 is still well below that level.

Anthropic Is Already Using Model 2 Internally

Model 2 is not simply a laboratory experiment. Anthropic says it and Mythos 5 are being used internally for coding, research, data generation and agent-based workloads.

The company is using the model while keeping it away from external users.

Anthropic says Model 2 has shown a noticeable improvement over Mythos 5 for many internal tasks, although the improvement is smaller than the jump seen in some of its previous model generations.

Why Isn't Model 2 Public?

The main issue is that Anthropic has not completed its usual full pre-deployment assessments for Model 2.

Because of that, the company says it has less confidence in its understanding of the model's capabilities than it does for models that have gone through the complete evaluation process.

Anthropic has also said that its internal review did not find new or more concerning forms of misalignment compared with those already observed in Mythos 5.

Model 2 Is Not Better at Everything

The benchmark results do not mean Model 2 is universally better than Mythos 5.

Other evaluations, particularly chemical and biological safety-related tests, showed Model 2 performing at levels comparable to or below Mythos 5. Those evaluations were also more limited.

Anthropic has warned that some of its existing evaluations are no longer keeping pace with improvements in AI capabilities, making it harder to produce a simple overall ranking between advanced models.

No Public Release Date Yet

For now, Anthropic has no plans to release Model 2 externally, and there is no confirmed indication that it will eventually launch under the name Mythos 6 or another commercial name.

That means users will have to continue using Anthropic's publicly available models while the company keeps Model 2 for internal work and further evaluation.