Google Gemini 4 Argon: Capabilities, Pricing and How It Compares

On October 13, 2026, Google released Gemini 4 Argon, the first member of the Gemini 4 family and, in Google's words, its "most capable model". Google says it is built for deep, multi-step work in coding, finance, law, cybersecurity and science. This article covers what Argon can do, where to use it, what it costs, and how it compares with the latest models from OpenAI and Anthropic, using only figures Google and the other companies have published.

What is Gemini 4 Argon?

Argon is Google's new flagship reasoning model. In the Gemini API it is called gemini-4-argon, and its default "thinking" level is medium. Google describes it as a model for complex software engineering, mathematical and scientific problem solving, autonomous multi-step agents, and long outputs. Its key technical feature is a unified 1 million token input-output limit in the Gemini API, which Google calls industry-leading, together with a new continuation protocol for very long tasks.

Google says Argon "is just the beginning" and that the rest of the Gemini 4 family will follow. For now, then, "Gemini 4" in practice means Argon.

Key capabilities

• Software engineering: According to Google, Argon in its Antigravity development platform increased agentic code submissions by more than 35% and reduced code rollbacks by nearly a third. It helped the Chrome team separate a roughly 6-million-line legacy folder from Chromium's codebase and delivered a 2.85x inference speedup for Gemma 4 E2B on vLLM.
• Professional work: Google says Argon follows style instructions better than previous Gemini models, hedges less and repeats itself less. It is first on the Vals Index, which measures performance in finance, coding, legal and tax work.
• Multimodal and 3D: stronger spatial reasoning and visual coding. Google showed demos of 3D scenes built in Blender and Three.js games.
• Science: Google cites contributions to quantum error correction, new counterexamples to Grad's conjecture in fusion research, rare-disease genome analysis and enzyme design.
• Robotics: 87% on the Embodied Reasoning benchmark (ERQA), which Google says beats all other frontier models.
• Cybersecurity: access for defenders through the Fairwind Program; Wiz used it in its Scan for Good initiative.

Safety

Google calls this its most comprehensive safety stack yet. Highlights: external bio red-teaming unlocked less than 1% of tested sub-tasks; "shortcut" attempts on coding benchmarks fell to 2%; and on the Gray Swan Indirect Prompt Injection benchmark, attack success across 15 adaptive attempts dropped to 0.7%. Argon was also trained to pause for human confirmation before irreversible steps.

Availability and pricing

Argon is available globally:

• Developers: the Gemini API paid tier in AI Studio and Android Studio. It is not available on the free tier.
• Enterprises: Gemini for business, the API and Google Antigravity.
• Consumers: the Gemini app and AI Mode in Search for Google AI Pro and Ultra subscribers. Antigravity access requires Ultra.

API pricing (per million tokens):

Item	Through Nov 30, 2026	From Dec 1, 2026
Input	$2	$4
Output (including thinking tokens)	$10	$20
Cached input (implicit caching)	$0.10	$0.20

Grounding with Google Search includes 5,000 free requests per month, then costs $14 per 1,000 requests.

Comparison with competitors: benchmarks

The table below comes from Google's official Argon model card. Note that these numbers are published by Google and not independently verified; each company picks the benchmarks that suit it.

Benchmark	Gemini 4 Argon	GPT-6 Astra	Claude Fable 5.1	Claude Opus 5.5
Vals Index (knowledge work)	68.9%	63.1%	65.8%	67.0%
AutomationBench	51.3%	41.4%	31.4%	42.5%
Vals Finance Agent v2	65.4%	53.5%	58.9%	58.6%
DeepSWE v1.1 (agentic coding)	77.9%	74.1%	67.4%	74.2%
FrontierSWE v2	55.0%	65.5%	56.3%	62.3%
Terminal-bench 4.0	57.4%	58.2%	57.9%	66.4%
Terminal-Bench Science 0.1	57.6%	68.1%	52.6%	63.3%
RiemannBench (math)	76.0%	72.0%	65.6%	69.6%
GraphWalks 256k-1M (long context)	84.2%	71.8%	65.0%	66.8%
OSWorld-2.0 (computer use)	69.2%	72.6%	—	—
LVBench (video)	91.7%	87.5%	79.7%	83.7%
CWE-bench v1 (security)	68.0%	68.0%	58.0%	67.0%

The pattern is clear: Argon leads in knowledge work, finance, very long context and video understanding. GPT-6 Astra is stronger on FrontierSWE, scientific terminal work and computer use, while Claude Opus 5.5 comes out on top on Terminal-bench 4.0.

Price comparison

Model	Release	Input / output (per 1M tokens)	Context
Gemini 4 Argon	Oct 13, 2026	$2 / $10 (intro), then $4 / $20	1M unified input-output
GPT-6 Astra	Sep 3, 2026	$10 / $50 (2x input, 1.5x output above 272K)	1.05M (128K max output)
Claude Opus 5.5	Sep 22, 2026	$4 / $20	1M (128K max output)

Even after the introductory period, Argon costs the same per token as Opus 5.5 and far less than GPT-6 Astra. Real costs also depend on how many tokens each model uses for a given task.

Conclusion

Gemini 4 Argon puts Google back at the front of the race, at least on its own benchmarks: strong in long, multi-step tasks, very long context and multimodal understanding, at an aggressive introductory price. It is not the best at everything, though. For terminal coding and some agentic tasks, Claude Opus 5.5 and GPT-6 Astra are still serious competitors. The best approach is to test all three on a sample of your real work before you choose, especially before prices double on December 1.

Note: models from other companies such as Meta, xAI and DeepSeek are not in the table because Google's official model card does not compare against them and we did not find comparable published figures.

Sources

• Google Blog — Gemini 4 Argon
• Google DeepMind — Gemini 4 Argon model card
• Gemini API pricing
• Gemini API — What's new in Gemini 4 Argon
• OpenAI — GPT-6 Astra
• OpenAI API docs — GPT-6 Astra
• Anthropic — Introducing Claude Opus 5.5
• Claude docs — Claude Opus 5.5

Cover image: Google (blog.google)