Gemini 4 Argon expands to Google AI Pro subscribers; API priced at $2/$10 per 1M tokens

Google is rolling out Gemini 4 Argon to Google AI Pro subscribers in the Gemini app and in AI Mode, less than a month after the model launched for AI Ultra subscribers only, according to 9to5Google. The model was announced at the end of September 2026.

Where Argon is available

Gemini app and AI Mode: AI Pro and AI Ultra subscribers now see four models in the picker: 3.5 Flash-Lite, 3.8 Flash, 3.1 Pro and 4 Argon. Both tiers can also use Argon in AI Mode. Ultra is listed at $99.99+ per month.

Developer and coding tools: Argon requires AI Ultra in the Antigravity desktop app, the Antigravity CLI, AI Studio and Android Studio (Canary).

Gemini API: Argon has an introductory price of $2 per 1M input tokens and $10 per 1M output tokens until November 30, 2026. After that it costs $4 per 1M input tokens and $20 per 1M output tokens.

App changes alongside the model

The Gemini app for macOS now lets users switch between chatting and coding, with direct access to Antigravity's agentic coding interface for building apps and web experiences. The main chat window also gets a new toggle for switching from chat to tasks, a feature previously called Spark.

Google claims Argon offers "frontier reasoning and improved performance across complex workflows." It also says the model:

• answers more directly with less hedging
• repeats itself less and avoids clichés
• follows style instructions "more reliably than any of our previous Gemini models"
• shares progress more often on longer tasks

These are Google's own characterizations. The company has not published independent verification in this announcement.

Google's internal usage claims

Google shared three examples from teams using Argon inside an internal version of Antigravity. All figures come from Google and have not been independently verified.

• Chrome: Google says Argon helped separate core webpage handling from a ~6 million line legacy folder of browser features inside Chromium's 40 million-line codebase. The refactor spanned hundreds of files, and Google says it finished "months of work in weeks."
• Gemma: Google claims a 2.85x end-to-end inference speedup on an A100 via vLLM for Gemma 4 E2B. Throughput went from ~37,900 to over 108,000 tokens per second, with what Google calls 100% token parity and no quality drop. Google says the upstream pull requests are public.
• YouTube: Google says Argon cut the CPU footprint of the comments frontend in half. It identified redundant emoji rendering as nearly 45% of server overhead and led to 12 targeted optimizations. Google reports a 30 ms drop in server comment-load latency and a 20 ms drop in client end-to-end latency.

Google is also offering 10,000 scientists the highest usage tier in Antigravity free for six months. Applications are open.

Not disclosed

The source material does not include Argon's context window, parameter count, training cutoff or benchmark scores. Those details are not yet confirmed here.

What this means

The move to AI Pro widens Argon's consumer reach well beyond the top-priced tier. Google is still holding its agentic coding surfaces (Antigravity desktop, CLI, AI Studio, Android Studio Canary) behind Ultra, which suggests it sees coding workflows as the premium differentiator.

The API pricing is the other signal. The introductory $2/$10 rate ends November 30, 2026, after which costs double to $4/$20. Teams evaluating Argon for production should budget at the higher rate.

The Chrome, Gemma and YouTube examples are measurements from Google's own engineers on Google's own codebases, so they say little about general performance. Without published benchmarks or a stated context window, independent testing will be needed to compare Argon against 3.1 Pro and competing frontier models.