Google released Gemini 4 Argon on Oct. 13, 2026, and paying consumers get it immediately: subscribers to Google AI Pro and Google AI Ultra can use it in the Gemini app and in AI Mode in Search. Developers face the other half of the launch. The API carries an introductory price of $2 per million input tokens and $10 per million output tokens, and after Nov. 30, 2026 that doubles to $4 and $20.
Koray Kavukcuoglu, Google DeepMind’s senior vice president and Chief AI Architect, wrote the announcement and called Argon Google’s latest frontier model. He wrote that “Gemini 4 Argon is just the beginning” and said more of the Gemini 4 family is on the way.
Pro and Ultra access, with Antigravity reserved for Ultra
Google’s launch post splits availability by audience. Google AI Ultra and Pro subscribers reach Argon in the Gemini app and AI Mode in Search, while Ultra subscribers alone get it in Google Antigravity. Developers and enterprises can call it through the API in AI Studio and the Gemini platform, and enterprises can also try it in Gemini for business. Google describes the global API release as available to paid-tier developers in AI Studio and Android Studio.
Free users are not mentioned in the post’s access list, so the consumer rollout as described starts and ends with the two subscription tiers. Anyone on a free account who wants Argon in the Gemini app would need to move up to Pro or Ultra, while the API route requires the paid tier in AI Studio.
The $2 and $10 introductory rate before the Nov. 30 doubling
Google’s announcement sets the introductory API price at $2 per million input tokens and $10 per million output tokens, with cached input tokens discounted 95%. From Nov. 30, 2026 the rates become $4 per million input tokens and $20 per million output tokens, and the post says the higher figure applies to the public-access Argon model. Current rates are posted on Google’s AI Studio pricing page. As an illustration of the arithmetic, a job that uses five million input tokens and one million output tokens costs $20 at the introductory rate and $40 afterward.
The window between launch and the change is about seven weeks, so any workload run before Nov. 30 bills at the lower rate while anything after it bills at exactly twice the price. The post gives no end date for the doubled rate and no separate price for any other Gemini 4 model, since only Argon is available so far.
Benchmarks, injection defenses and the scientist program
Google reports that Argon ranks first on the Vals Index, which measures finance, coding, legal and tax work, and scores 87% on the ERQA embodied-reasoning test. Two results come from private evaluations run by Gamow Labs: 57% on RareBench, which covers help with rare-disease diagnosis, and 52.2% on LabBench wet-lab reasoning. On safety, the post says violations on an adversarial set of harmful requests fell nearly twofold against earlier Gemini models, and shortcut attempts on coding benchmarks dropped to 2%. The full list of evaluations is in Google’s model card.
The prompt-injection figure is the most specific: through adversarial training and automated red teaming, Argon leads the Gray Swan Indirect Prompt Injection benchmark, with attack success cut to 0.7% across 15 adaptive attempts. Indirect prompt injection means hidden instructions planted in content a model reads, such as a web page or document, and the benchmark measures how often an attacker’s repeated, adjusted tries succeed in steering the model.
Google is also offering 10,000 scientists the highest usage tier in Antigravity free for six months, with onboarding that is rolling and subject to eligibility, capacity and selection. Separately, the company is expanding its Fairwind Program to hundreds of additional partners, and approved cyber defenders can use Argon without cyber guardrails on a rolling basis.
Chrome refactors, Gemma speedups and a humanoid’s cornhole shots
The announcement leans on examples from inside Google. Argon helped finish an architectural refactor across hundreds of files in Chrome that the team had estimated would take months, completing it in weeks. In the Agent Development Kit, optimizations cut in-memory session preparation time by 40% in long sessions and sped up cold imports by 21%, work described in an engineering write-up. Quantum-chemistry code xTB ran up to 2.3 times faster at predicting molecular structures, and Gemma inference on an A100 reached a 2.85 times end-to-end speedup, with throughput rising from about 37,900 to more than 108,000 tokens per second and no reported quality drop.
Robotics gets a lighter example. Argon learned a cornhole skill from video, and the transferred skill let Apptronik’s Apollo humanoid sink seven consecutive shots in a real-world trial, per a developer post. Every figure above is Google’s own reporting, and the announcement offers no outside replication, so the Vals Index ranking and the speedups stay unverified until independent testers publish results.
This article was produced with the assistance of AI and reviewed by Morning Overview editors prior to publication.
More from Morning Overview
- Hybrids have 15% fewer problems than gas cars, while EVs and plug-in hybrids have about 80% more, Consumer Reports finds
- Regulators cleared the first U.S. small modular reactor, 4 months early
- Lake Powell sank to a record 3,517 feet, nearing the level that stops Glen Canyon Dam’s turbines
- Seven of Earth’s nine planetary boundaries are breached, and all seven are worsening
