Google prices its third-ranked Gemini 4 Argon at half the cost of Anthropic's Claude Opus 5.5

Take a job that reads a million tokens and writes a million. Argon costs $12 against $24 on Claude Opus 5.5 and $60 on Claude Fable 5.1, using the list rates Jefferies cited ^[1][18][15]. Google prices cached input 95% lower, about 10 cents per million tokens ^[13].

The output ceiling moved as well. Argon can write up to 1 million tokens in a single trajectory, up from 64,000, about 15.6 times the old limit ^[6][17]. A per-token discount saves the most on long jobs, and long jobs are what Tim Law, director of research for AI at IDC, singled out. "Google Gemini 4 Argon shows advanced reasoning on critical tasks, according to benchmarks, particularly legal reasoning, finance and other aspects of enterprise knowledge work, including long-running tasks," Law said ^[21].

Jefferies said Argon appears to set a new benchmark for the cost-performance frontier ^[18]. Against Anthropic the case is price. Against OpenAI, whose discounted GPT-6.1 Sol carries the same rates ^[11], the case is rank (only the two Claude models sit above Argon on the Artificial Analysis index, so every OpenAI model on that board sits below it) ^[25].

Other outcomes are possible. Sonnet 5.5 also ranks above Argon, and its price is not in any of the three reports ^[24]. If it sells near $2 and $10, Google's discount argument against Anthropic shrinks to Opus alone. Anthropic could also cut Opus 5.5 before Google opens access. Or production results could trail the benchmarks, a risk Law named himself. "While Google touts the success of internal use cases, the final proof will be in enterprise production environments once the model is fully released," he said ^[10]. Jefferies, for its part, called independent testing and real-world deployment the "ultimate validation" ^[9].

Before wider release, Google is going through the Trump administration's voluntary pre-release model access process ^[5]. "Starting this rollout in this way gives us more confidence, but also enables us to put a model that is trained and strong in cyber defense in the hands of defenders as soon as possible," said Tulsee Doshi, Google's Gemini model product lead ^[8]. So far the only payback Google has described is internal. Argon agents found memory optimizations that freed more than 300 TiB across its data centers, with total savings estimated at 500 TiB to 1 PiB, and Google did not have to buy additional hardware to get them ^[3][4].

I think Google priced its first flagship since February ^[16] to win enterprise trials on cost while it trails on rank. Its own plan keeps Argon for the hard work. Doshi said the cheaper, lighter models Google released over the summer remain essential for everyday agents, with frontier models handling complex assignments ^[23]. The counter-case is that enterprise buyers pay for the top-ranked model, and if so, halving the price buys Google little. The view is wrong if Sonnet 5.5 already sells near Argon's rates, or if Anthropic halves Opus 5.5 before Google opens the API.