Google has officially launched Gemini 4 Argon, its newest frontier AI model, releasing it to developers and enterprises worldwide on October 13 at prices positioned below competing offerings from OpenAI and Anthropic as the company moves to win more business and developer customers.

Pricing designed to undercut rivals

According to Google's launch blog, introductory pricing is set at $2 per million input tokens and $10 per million output tokens, with cached inputs — repeated context such as system prompts — costing 95% less, $0.10 per million tokens. The rates remain in effect until November 30, after which they double, according to the blog post.

That entry point sits below both main competitors: Anthropic's Claude Opus 5.5 charges $4 and $20 per million input and output tokens respectively, while OpenAI's GPT-6 Astra charges $10 and $50. Measured on a per-task basis, Mashable puts Argon at about $1.99 against the Artificial Analysis Intelligence Index, which is almost 60% of what GPT-6 Astra costs. Because frontier APIs bill per million tokens — the text fragments models read and generate — those list-price gaps compound directly for high-volume applications.

Developers and enterprises will assess Argon through the Gemini API in AI Studio, Google's developer environment for building with Gemini models, and the Gemini platform. Google AI Pro and Ultra subscribers can use it directly in the Gemini app and AI Mode in Search, while the Antigravity coding tool remains limited to Ultra accounts, according to Google's post on X:

Gemini 4 Argon is available now for Google AI Ultra and Pro subscribers in @GeminiApp and AI Mode in Search, Ultra subscribers in @Antigravity, and to developers and enterprises via the Gemini API in @GoogleAIStudio and the Gemini platform. Argon will launch at an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off input token price.

— Google (@Google) October 13, 2026

Where Gemini Argon lands on benchmarks

Google says Argon tops the Vals Index, an evaluation that measures performance across finance, coding, legal and tax tasks. The company's own comparison table, however, shows Argon leading in 14 of 18 results, tying with GPT-6 on cybersecurity and trailing in five.

On Artificial Analysis' Intelligence Index, which tests models on standardized hardware, Argon scores 53 on its “High” setting. That ties with GPT-6 Astra and just edges out GPT-6.1 Sol and Meta's new Muse Wonder Max by a point. It trails Claude Opus 5.5 and Claude Sonnet 5.5 by about five points, while still marking a jump of 23 points over Gemini 3.1 Pro Preview, Google's previous top model. For evaluators, that sets up a direct trade-off: a roughly five-point deficit to Anthropic's models on the index, against an output rate half of Opus 5.5's $20 per million tokens.

Early customers and a cyber carve-out

Google named consulting firm Deloitte, legal research provider LexisNexis, healthcare technology company Ambience Healthcare, and drug discovery company Isomorphic Labs among the first firms testing Argon. The group spans consulting, legal research, healthcare and drug discovery — the categories Google chose to spotlight at launch. The tech giant is also expanding its Fairwind Program to include hundreds more partners, giving vetted cybersecurity defenders access to a version of the model with its cyber safeguards removed.

Google claims its own engineering teams running Argon in Antigravity increased agentic code submissions by more than 35% and reduced code rollbacks by almost a third. The company also credited the model with helping Chrome engineers complete an architectural overhaul spanning hundreds of files in Chromium's 40-million-line codebase.

Launch follows a rocky period at DeepMind

Google stated in September that it would no longer ship Gemini 3.5 Pro, a model CEO Sundar Pichai had once slated for June. The company later reorganized Google DeepMind, its AI research division, after co-founder Demis Hassabis stepped aside and several Gemini leaders left. Argon first surfaced on September 30 in a post by DeepMind SVP Koray Kavukcuoglu, who said access was initially limited to vetted cyber defenders in the Fairwind Program.

Google says Argon rejects requests involving cyberattacks or chemical, biological, radiological and nuclear threats. On Gray Swan's prompt-injection test, it recorded a 0.7% attack success rate, compared with 1.0% for Opus 5.5 and 8.5% for GPT-6 Astra, with lower scores indicating stronger resistance.

Attention now turns to November 30, when Argon's introductory API pricing is set to expire and its cost advantage over rivals begins to narrow.