Half price until 30 November

Google on Tuesday made Gemini 4 Argon generally available to developers and enterprises through its API, at an introductory price of $2 per million input tokens and $10 per million output tokens, according to a post by Koray Kavukcuoglu, senior vice-president of Google DeepMind and Google's chief AI architect. After the introductory period ends on 30 November, the price becomes $4 and $20, Google said. By our arithmetic that is exactly double on both sides, so the launch rate is a promotion with a stated end date.

Google first unveiled Argon on 30 September, when access was limited to a small group of early testers, VentureBeat reported. Tuesday's release opens it through the paid Gemini API tier in AI Studio and on Google's enterprise platform. Individual developers need a Google AI Ultra subscription to use it in Antigravity, Google's development environment, and Ultra and Pro subscribers get it in the Gemini app and in AI Mode in Search. Cached input tokens are 95 per cent cheaper during the introductory period, Google said.

Against list prices published by the other developers, the introductory rate matches the $2 and $10 that Anthropic charges for Claude Sonnet 5.5 and that OpenAI charges for GPT-6.1 Sol. The post-promotion rate of $4 and $20 matches Anthropic's Claude Opus 5.5 and Meta's Muse Wonder. OpenAI's GPT-6 Astra and Anthropic's Claude Fable 5.1 are listed at $10 and $50. VentureBeat, which compiled a similar table, noted that per-token prices exclude the cost of tools, infrastructure and other services.

Google's post says Argon ranks first on the Vals Index, which measures performance on finance, coding, legal and tax work, and that it scored 87 per cent on ERQA, an embodied-reasoning benchmark. It says Argon outperforms other frontier models without naming them. VentureBeat reported the Vals figures as of 12 October: Argon at 68.90 per cent, Claude Sonnet 5.5 at 67.04 and Claude Opus 5.5 at 66.97, a lead of about 1.9 points by our subtraction.

Other measures rank the models differently. On the Artificial Analysis Intelligence Index, VentureBeat reported, Argon scores 53, level with GPT-6 Astra and Claude Fable 5.1 and below Claude Opus 5.5, Claude Sonnet 5.5 and Meta's Muse Wonder Full. On Terminal-Bench 4.0, a test of coding through command-line tools, figures Google had disclosed earlier put Opus 5.5 at 66.4 per cent and Argon at 57.4, VentureBeat said, describing them as company-reported rather than independently replicated.

Google also cited results from its own use. Argon in Antigravity helped its engineers increase agentic code submissions by more than 35 per cent while reducing code rollbacks by nearly a third, the post says, and cut in-memory session preparation time in Google's Agent Development Kit by 40 per cent in long sessions. Google says Argon leads the Gray Swan indirect prompt-injection benchmark, with attack success held to 0.7 per cent across 15 adaptive attempts. These are Google's figures.

Access and safety terms are part of the release. Google said it is beginning to extend the Fairwind Program, which gives approved cybersecurity defenders access to Argon without its cyber guardrails, to hundreds of additional partners, with Wiz among the early users. It said external biological red teaming could unlock less than 1 per cent of the biological sub-tasks it tests for. The model offers a one-million-token combined input and output limit through the Gemini API, according to the post; VentureBeat noted that Google had not said how that limit relates to the length of a single response.

Why it matters is the shape of the price rather than its level. A temporary rate that matches rivals' cheaper models sets a reference point that developers will budget around, and the doubling takes effect on 1 December. VentureBeat's reading is that Google's advantage lies in the breadth of products and infrastructure through which it can distribute Argon rather than in any single benchmark.

Komposite reported in September that frontier list prices fell when Anthropic and OpenAI released new models, and that McKinsey found operating costs still constrain AI use at one in five organisations. Per-token price is only part of the bill.