Google and Anthropic release models with different access and pricing rules

The naming and access gap

There is significant confusion regarding the current flagship models from Google and Anthropic. While many comparisons mention 'Gemini 4 Pro,' Google has not actually released a model by that name. The production model is actually Gemini 4 Argon, which was opened to the public in October 2026.

In contrast, Anthropic's Claude Mythos 5 is not a general-purpose product available to everyone. It remains a gated specialist that cannot be purchased with a standard credit card. Instead, users must go through a formal application process to gain access to the model.

As OrcaRouter reported, the primary difference between these two is that one is an open generalist and the other is a restricted specialist. This distinction means that a direct performance comparison is often less relevant than the ability to actually use the software for a specific project.

Gemini 4 Argon capabilities

The most notable technical change in Gemini 4 Argon is its output ceiling. It can generate up to one million tokens in a single response, which is a significant increase from the 64K limit of the previous generation. This allows the model to produce long documents or entire test suites in one go without needing to stitch pieces together.

Google has reported several benchmarks for the model, including a first-place finish on the Zapier AutomationBench. It also performed well on long-video understanding and vulnerability remediation tasks. However, these figures are vendor-reported and have not been independently reproduced by third parties.

While the model excels in several areas, it does show some weaknesses in specific categories. It trails behind other models like Claude Opus 5.5 and GPT-6 Astra in certain science and terminal-based benchmarks. These results suggest that while Argon is powerful, it has specific strengths and limitations depending on the task.

Pricing and cache dynamics

The pricing for Gemini 4 Argon is currently on an introductory promotion that expires on November 30, 2026. After that date, the price will increase to $4 per million input tokens and $20 per million output tokens. Many current comparison pages fail to account for this upcoming price hike when evaluating the cost gap.

Another critical factor for developers is the cost of cache reads. Gemini 4 Argon offers a 95% discount on cache reads, bringing the cost down to $0.10 per million. This is significantly cheaper than Claude Mythos 5, which charges $1.00 per million for the same operation. This difference matters for agents that frequently re-read large datasets.

Claude Mythos 5 does have one pricing advantage for specific use cases. It offers batch requests at half price, which can be beneficial for non-urgent processing. However, for most real-time applications, the high output price and the lack of deep cache discounts make it a much more expensive option than the Google alternative.

Security and restrictions

Anthropic has placed strict rules on the use of the Mythos 5 model to ensure it is used by authorized organizations. It is part of a three-tier program that includes Defense Access, Red Team Access, and a tier for critical infrastructure. These tiers require a review process that is currently limited to US organizations at launch.

A major hurdle for many businesses is the mandatory 30-day data retention policy for all Mythos-class traffic. This policy automatically rules out any organization that requires a zero-retention commitment for their data. Because of these rules, the model is effectively hidden from the general public and most independent evaluation pipelines.

Because Mythos 5 is so restricted, it does not appear on many neutral intelligence indices. This absence of a score is a key piece of information for researchers. It confirms that the model is being treated as a specialized tool rather than a general-purpose AI that can be compared side-by-side with standard offerings.

Choosing the right model

The decision between these models depends entirely on the nature of the work being performed. If a team is doing authorized offensive security or malware analysis and can pass the review, Claude Mythos 5 is the intended specialist. In this scenario, the application process is simply the cost of admission for the model's specific capabilities.

For teams focused on long-output code generation or document synthesis, Gemini 4 Argon is the more practical choice. It is available without an application and is currently priced at a promotional rate. It provides a high output ceiling that allows for complex tasks to be completed in a single trajectory without extra steps.

Users who fall between these two categories might find success with Claude Fable 5.1. This model provides the same underlying technology as Mythos 5 but with standard safeguards and no verification program. Testing Fable 5.1 first can help a team determine if it meets their needs before they commit to the lengthy review process for the unrestricted Mythos model.