Artificial Intelligence
Google rolls out Gemini 4 Argon
Google introduced Gemini 4 Argon as its latest frontier model, set a staged release and token pricing, and early benchmark results placed it level with some rivals but behind others.
- Google announced Gemini 4 Argon and said it could take text, images, video and audio as input while producing up to one million output tokens.
- The company said it would first release the model to trusted cyber defenders and its own internal teams before widening access later.
- Google set an introductory price of $2 per million input tokens and $10 per million output tokens, with cached inputs discounted by 95 percent.
- Artificial Analysis scored Argon at 53, level with GPT-6 Astra (max) and Claude Fable 5.1, and below Claude Opus 5.5 at 58.
- Vals AI placed Argon at 68.9 percent on its index, and Arena.ai ranked it first in text.
Google unveiled Gemini 4 Argon as its latest frontier model, following Gemini 3.1 Pro after more than seven months. It said the system accepted text, images, video and audio as inputs and could generate up to one million output tokens.
The first access went to trusted cyber defenders and Google's internal teams, with the model released without cyber guardrails. Google said the early phase would help shape the safety systems before wider access to API customers, Google AI Ultra subscribers, developers, businesses and consumers.
The introductory price was set at $2 per million input tokens and $10 per million output tokens, with cached inputs priced 95 percent lower. Google said the standard rate would later rise to $4 per million input tokens and $20 per million output tokens.
Artificial Analysis gave Argon 53 points on its Intelligence Index, matching OpenAI's GPT-6 Astra (max) and Anthropic's Claude Fable 5.1. The same assessment put Claude Opus 5.5 ahead on 58 points and said Argon used more output tokens per task than GPT-6 Astra.
Google said its own benchmark results put Argon ahead in most of the tests it disclosed. Vals AI ranked it first on its index, and Arena.ai placed it first in text and among the strongest models across several writing-related categories.
Named in this story
Companies
- announced Gemini 4 Argon and set its rollout and pricing
- OpenAI
- rival whose models were used in the comparisons
- Anthropic
- rival whose models led several of the benchmarks
- Vals AI
- ran the Vals Index where Argon ranked first
- Arena.ai
- hosted the human preference rankings
Organisations
- Artificial Analysis
- published the independent benchmark assessment
Governments and agencies
- US government
- ran the voluntary pre-release access programme
Products and systems
- Gemini 4 Argon
- Google's new frontier model
- GPT-6 Astra
- OpenAI model that matched Argon on one index and used fewer tokens
- Claude Opus 5.5
- Anthropic model that scored higher than Argon on Artificial Analysis
- Claude Sonnet 5.5
- Anthropic model that outscored Argon on some tests
- Claude Fable 5.1
- model that tied Argon on Artificial Analysis
- GPT-6.1 Sol
- OpenAI model that Argon edged on the Artificial Analysis index
- Gemini 3.1 Pro
- Google's previous frontier model
- Gemini 3.8 Flash
- earlier Google model used as a comparison point
How the source tells it
The source read like a launch piece with strong boosterish and leaderboard-driven framing, although it still rested on benchmark scores and rollout details.
- vendor boosterism the release was framed through Google's own claims about closing the gap and moving back among the leaders
- competitive triumph benchmark tables and first-place language turned the announcement into a ranking contest
- novelty hype the one million-token output limit and industry-first wording presented a configuration change as a breakthrough