Google’s Gemini 4 Argon has matched OpenAI’s GPT-6 Astra on the Artificial Analysis Intelligence Index, scoring 53 points. At current launch prices, it also costs less to run per test task.

Google DeepMind announced the model on September 30. For now, only a group of trusted cyber defenders can use it.

Introducing Gemini 4 Argon – our new frontier model.It’s built for complex workflows across coding, enterprise knowledge work, and cybersecurity defense – rolling out today to a set of trusted testers through our Fairwind Program. pic.twitter.com/X8acOWJOSF

— Google DeepMind (@GoogleDeepMind) September 30, 2026

Argon Puts Google Back Among the Top 3 Labs

Artificial Analysis, an independent benchmarking firm, said the score puts Google back among the top 3 AI labs. Argon also sits 1 point above OpenAI’s GPT-6.1 Sol, which scored 52.

The model scored 23 points higher than Gemini 3.1 Pro Preview, Google’s previous non-Flash model. Earlier in September, Google released Gemini 3.8 Flash alongside a cybersecurity variant. For comparison, SpaceXAI’s Grok 4.7 scored 46 on the same index in September.

Artificial Analysis credited Argon’s gains to fewer hallucinations and stronger agentic skills. On its AA-Omniscience test, Argon posted a 15% hallucination rate, compared with 51% for Astra. However, Argon answered 50% of those questions correctly, 13 points below Astra.

In agentic work, Argon led AutomationBench-AA at 77.5%, ahead of Anthropic’s Claude Sonnet 5.5 at 71.3%. Still, it trailed Sonnet 5.5, Claude Opus 5.5, and Astra on Terminal Bench 4.

Follow us on X to get the latest news as it happens

The Smaller Bill Runs on a Launch Discount

Google has set introductory pricing at $2 per million input tokens and $10 per million output tokens. Those rates double to $4 and $20 once the promotion ends.

At the discounted prices, Artificial Analysis measured Argon’s cost at $1.99 per Index task, against $3.26 for Astra. However, at standard rates, the figure rises to $3.98, about 1.2 times Astra’s cost.

The savings come from cheaper tokens, since Argon uses more of them. It averaged 62,000 output tokens per task, compared with 27,000 for Astra. The model also has a 1 million token output limit.

“To support Gemini 4 Argon’s capabilities across longer, more complex use cases, we are significantly expanding the model’s output token limit to an industry-leading 1M tokens, up from the previous 64K tokens,” the team said.

Cyber Defenders Get the First Look at Gemini 4 Argon

Google CEO Sundar Pichai said teams across Google already use it heavily, with work ranging from coding to quantum computing.

“Importantly, Argon has frontier safeguards, and we are rolling it out responsibly – it’s with the US gov’t and going to a set of trusted cyber defenders through our Fairwind Program today,” he noted.

Google’s launch post says Argon can find, validate, and patch serious software flaws on its own. For that reason, trusted defenders and Google’s internal teams will receive it without cyber guardrails. 

Paid API customers and Google AI Ultra subscribers will get access first when the rollout widens. Google has not said whether the launch discount will still apply by then.

The firm also said it is engaged in the US government’s voluntary pre-release access process. 

Subscribe to our YouTube channel to watch leaders and journalists provide expert insights