Google opened Gemini 4 Argon, its new frontier model, to the public on October 13, 2026, ending a two-week period in which only vetted cybersecurity defenders could use it. Developers can now call the model through the Gemini API's paid tier inside Google AI Studio, enterprises can reach it on Google Cloud, and Google AI Pro and Ultra subscribers get it in the Gemini app and in AI Mode in Search. Access inside Antigravity, Google's agent-first coding environment, is reserved for Ultra subscribers. The Gemini API changelog lists gemini-4-argon as generally available from the same date.
Pricing is the most aggressive part of the launch. For an introductory period Google is asking $2 for every million input tokens and $10 for every million output tokens, and cached input costs 95 percent less. The Verge reports that those rates will double when the introductory period ends in late November, and estimates that even the launch price is roughly a fifth of what OpenAI and Anthropic charge for their top models. Per-token prices do not translate directly into task costs, since models can consume very different numbers of tokens for the same job.
Google pitches Argon as a model for long, multi-step professional work in coding, finance, law, cybersecurity and science. The API exposes a unified limit of one million tokens shared between input and output. Most of the evidence offered is internal. Google says engineers using Argon in Antigravity raised agentic code submissions by more than 35 percent while cutting rollbacks by nearly a third. It also describes Argon helping the Chrome team split a legacy folder of roughly six million lines out of Chromium's 40-million-line codebase, trimming in-memory session preparation in the open-source Agent Development Kit by 40 percent, and nearly tripling inference throughput for the small Gemma 4 E2B model on an A100 GPU.
On benchmarks, Google says Argon ranks first on the Vals Index, which scores models on finance, legal, coding and tax tasks, and VentureBeat confirms that placement. The picture from broader testing is less one-sided. Artificial Analysis, as cited by VentureBeat, puts Argon at 53 on its Intelligence Index. That ties it with GPT-6 Astra from OpenAI and Claude Fable 5.1 from Anthropic, while Anthropic's Claude Opus 5.5 and Claude Sonnet 5.5, along with Muse Wonder Full from Meta, score higher. Google's own earlier comparisons also showed Opus 5.5 ahead on Terminal-Bench 4.0, 66.4 percent against 57.4 percent. Sam Witteveen of Red Dragon AI, who tested the model early, told VentureBeat it was very strong but not, in his view, the best available.
Safety framed the staged rollout. Google first released Argon on September 30 only to partners in its Fairwind Program and, according to VentureBeat, took part in a voluntary US government pre-release access process. A Google DeepMind post says the model solved all 12 of its end-to-end simulated cyberattack scenarios, compared with five for the previous generation, and showed elevated capability in biology. Google says it responded with layered safeguards under its Frontier Safety Framework: external biology red teamers unlocked under 1 percent of tested sub-tasks, shortcut attempts on coding tasks fell to 2 percent, and attack success on the Gray Swan indirect prompt-injection benchmark dropped to 0.7 percent. Fairwind is now widening to hundreds of additional defenders.
The release also carries robotics and research claims. Google reports an 87 percent score on the ERQA embodied-reasoning benchmark and describes a test in which Argon turned a video of a person playing cornhole into simulation code that an Apptronik Apollo humanoid then used to land seven throws in a row. It is offering 10,000 scientists six months of the top Antigravity tier at no cost, subject to eligibility and capacity.
For Google, the stakes are competitive. The Verge notes that Argon is the first big frontier model Google has shipped since Gemini 3 arrived in late 2025, a stretch in which Anthropic and OpenAI pulled ahead, and that it arrived a day after Meta's Muse Wonder Full. Low introductory pricing and distribution across Search, the Gemini app and Google Cloud give Argon a broad route to users, but most of its performance and productivity figures are still Google's own and await independent replication.