Google unveiled Gemini 4 Argon as its new frontier model. The release closes the gap with rivals from OpenAI and Anthropic while beating some of them on key benchmarks. Argon is Google's first frontier model in more than seven months, following Gemini 3.1 Pro. The launch returns the ad giant among the top three AI labs, though Anthropic likely still holds the lead. Google skipped the already-announced Gemini 3.5 frontier model entirely after a difficult and drawn-out development period.
Initial Rollout and Phased Safety Approach
Argon is initially going to a group of trusted cyber defenders as part of the Fairwind program. Those defenders and Google's internal teams receive the model without cyber guardrails. Google justifies the gradual rollout with a phased approach that AI capabilities at this level require. The company is also taking part in the US government's voluntary program that gives agencies access to new models before public release. Feedback from early testers will feed into the safety mechanisms.
Only after that phase does Google plan to open Argon up to developers, businesses, and consumers. The rollout starts with paying API customers and Google AI Ultra subscribers. The company has not given an exact date, saying only as soon as possible. Pricing is already set at an introductory rate of $2 per million input tokens and $10 per million output tokens. Cached input tokens cost 95 percent less, working out to about 10 cents per million.
Benchmark Scores and Pricing Metrics
Google also raised the output limit from 64,000 to one million tokens, which the company calls an industry first. Argon accepts text, images, video, and audio as input but only outputs text. Artificial Analysis provides an early independent assessment showing Argon scores 53 points on its Intelligence Index at the highest reasoning level called High. That ties the model with OpenAI's GPT-6 Astra max and Claude Fable 5.1, placing it one point ahead of GPT-6.1 Sol max. Anthropic's models still lead with Claude Opus 5.5 at 58 points and Claude Sonnet 5.5 at 56.
At the current promo price, one Intelligence Index task costs $1.99. That figure is 60 percent of GPT-6 Astra's cost at $3.26 but 2.7 times more expensive than GPT-6.1 Sol. Once the discount ends, the cost rises to $3.98, which is about 20 percent above GPT-6 Astra. The price advantage comes from lower token rates rather than efficiency because Argon uses an average of 62,000 output tokens per task while GPT-6 Astra needs only 27,000.
Read nextGoogle Shuts Down Gemini Gems and Replaces Them with SkillsAgentic Tasks and Human Evaluation Rankings
Argon made significant gains on agentic tasks, which have historically been a weak spot for Gemini models. On AutomationBench-AA, the Artificial Analysis variant, Argon takes first place at 77.5 percent, six points ahead of Claude Sonnet 5.5 max. On Terminal Bench 4, it hits 57 percent, marking a 53-point jump over Gemini 3.1 Pro Preview. That leaves it behind Claude Sonnet 5.5 at 64 percent, Claude Opus 5.5 at 60 percent, and GPT-6 Astra at 59 percent. Artificial Analysis also highlights Argon's low hallucination rate of 15 percent.
On Arena.ai, where humans rate model outputs in head-to-head comparisons, Argon performs well. In the Text Arena, Gemini 4 Argon High takes first place with 1,525 points, putting it 20 points ahead of Claude Opus 4.6 High in second. According to Arena, Argon leads in coding, hard prompts, instruction following, longer queries, and creative writing. Web development results remain more modest, with Argon scoring 1,679 points to land in eighth place in Code Arena WebDev. On price-to-performance, Arena puts Argon ahead of the field as the most cost-efficient model in the ranking.
