Gemini 4 Argon debuts with a unified 1M input/output token limit
The New Stack Frederic Lardinois ● Covered by 8 sources
Google opened Gemini 4 Argon to developers with a shared 1M token budget. That means long prompts can now crowd out the answer, even as Google prices it to look cheap.
Based on reporting by The New Stack, Frederic Lardinois — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Google has opened Gemini 4 Argon to developers on the paid Gemini API tier, via Google AI Studio and several other tools. The model was announced on September 30, but early access was limited to a small set of cyber defenders in Fairwind and other trusted testers, with Google employees using it internally first.
The pricing is meant to turn heads. Through November 30, developers pay $2 per million input tokens and $10 per million output tokens, with a single 1 million-token pool shared between input and output. That matches OpenAI’s GPT-6.1 Sol on price. After the introductory period, Google says Argon moves to $4 and $20 per million tokens, which lines it up with Anthropic’s Claude Opus 5.5. Cached input is 95 percent off during the promo window.
But the token setup is the real twist. Google originally framed Argon as having a 1 million-token output cap. Now it describes the limit as a combined budget for what goes in and what comes out. So a huge prompt can eat a big chunk of the room left for the reply. Google says that tradeoff should help with long coding jobs, science work, and agent tasks. Anthropic’s Opus 5.5, by contrast, keeps a 1 million-token context window and a separate 128,000-token output cap.
On benchmarks, Google is making a strong case. It says Argon ranks first on the Vals Index, which weighs finance, coding, legal, and tax work by their share of U.S. GDP. On ERQA, a test of physical and spatial reasoning, Google reports 87 percent, ahead of Claude Opus 5.5 at 82.3 percent, GPT-6 Astra at 81 percent, and Claude Fable 5.1 at 74 percent. Bloomberg had reported that some employees found the model shaky on certain coding tasks, especially front-end design, and Google pushed back on that characterization.
Availability is wide, but not uniform. Argon is in the Antigravity desktop app for macOS, Windows, and Linux, and in the Antigravity CLI, but only for Google AI Ultra subscribers and eligible Gemini Enterprise plans. AI Pro subscribers get it in the Gemini app, not Antigravity. Google is also putting Antigravity’s coding interface into the Gemini app for macOS, and subscribers on AI Pro and Ultra can use Argon in the Gemini app and in AI Mode in Search.
My take — AI-written commentary, not fact-checked reporting
Google keeps trying to sell frontier AI as both cheaper and safer, which is the usual corporate magic trick: lower the sticker price, raise the confidence. The shared token budget is the more honest part here, because it admits long prompts and long answers are fighting for the same space. That’s the sort of detail that matters more than another glossy demo of a wizard chess game.
Read more about this at: The New Stack