Google releases Gemini 4 Argon, its new frontier model: Benchmarks, how to try it, and pricing

Google announced the public launch of Gemini 4 Argon after an earlier limited release to select testing partners. It’s the first model in the company’s Gemini 4 generation and now sits at the top of Google’s lineup.

Google CEO Sundar Pichai announced the launch on X, where he said the model would be available globally starting today.

This Tweet is currently unavailable. It might be loading or has been removed.

Google first announced Argon on Sept. 30 in a blog post by Google DeepMind SVP Koray Kavukcuoglu. At that point, it was limited to trusted cyber defenders in Google’s Fairwind Program. With today’s release, Google AI Pro and Ultra subscribers can use Argon in the Gemini app.

Google says Argon is built for long, complex work in software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense. For everyday chat, Google says the model gets to the point with less hedging, repeats itself less, and avoids clichés. On longer tasks, it gives more frequent progress updates.

Argon 4 arrives during a busy fall season for frontier AI models. Google repeatedly missed its own deadlines for the launch of Gemini 3.5 Pro, and meanwhile, Anthropic released Claude Opus 5.5, and OpenAI released GPT-6 Astra. Meta’s “super intelligence” Muse Wonder was just released yesterday.

As expected, Google’s own evaluation shows Argon testing ahead of its rivals on many benchmarks. On the independent Artificial Analysis Intelligence Index, Argon ties GPT-6 Astra at 53 with high reasoning enabled, one point above GPT-6.1 Sol and the new Muse Wonder Max. The same index places it about five points behind Claude Opus 5.5 and Claude Sonnet 5.5.

Coding is the clearest focus of the release. Google says Argon already handles much of its internal engineering work, and today’s update brings Google Antigravity’s agentic coding interface directly into the Gemini app on macOS.

Want to learn more about getting the best out of your tech? Sign up for Mashable’s Top Stories and Deals newsletters today.

Google Gemini Argon 4: How to try it

Starting today, Google AI Pro and Ultra subscribers can use Gemini 4 Argon in the Gemini app. Google AI Pro costs $19.99 a month. At I/O 2026, Google added a $100-a-month Ultra plan and lowered its top Ultra tier from $250 to $200 a month.

Google set an introductory API price for Argon:

• $2 per million input tokens

• $10 per million output tokens

• Cached input tokens at 95 percent off, or $0.10 per million

That is less than both main rivals charge. Anthropic’s Claude Opus 5.5 costs $4 and $20 per million input and output tokens, while GPT-6 Astra costs $10 and $50 per million tokens. Today’s announcement focuses on the Gemini app and doesn’t say whether API access has widened beyond Google’s earlier rollout plans.

The rollout comes with several other changes to the Gemini app. First, Antigravity’s agentic coding interface now lives inside the Gemini app on macOS, so users can move between chatting and coding without switching apps. Second, a new toggle in the main chat window switches Gemini from chat to tasks. In tasks mode, Gemini can keep working across your Google apps and other services while you’re away from your computer. Pro and Ultra subscribers get access.

Additionally, Gemini now connects to more than 40 apps, including Instacart, OpenTable, Adobe, and Shopify. Google pitches this for jobs like ordering groceries, booking restaurant tables, and designing or building an online business.

Google Gemini Argon 4: Benchmark performance

Google’s comparison table has 18 results. Argon has the top score on 14, ties with GPT-6 in cybersecurity, and trails on five. All figures in this section come from Google. Typically, AI companies release results that show their own model scoring above its competitors.

Argon’s biggest leads are in knowledge work. Google says it tops the Vals Index, which weights finance, coding, legal, and tax work by each sector’s share of U.S. GDP. On AutomationBench, Zapier’s test of end-to-end business tasks, Argon scored 51.3 percent. That is about nine points ahead of the next model.

The coding results are mixed. Argon set a new high on DeepSWE v1.1, Google’s main test of long software engineering tasks, at 77.9 percent. But it trails GPT-6 Astra on FrontierSWE v2. On Terminal-Bench 4.0, it trails all three rivals, and Opus 5.5 leads.

Some of Google’s figures for competitors don’t match what Anthropic reported. Mashable reported that Anthropic put GPT-6 Astra at 57.9 percent on Terminal-Bench 4.0, while Google lists 58.2 percent. On the Terminal-Bench Science benchmark, Anthropic reported 58.7 percent for Opus 5.5 (default effort), while Google lists 63.3 percent. When Anthropic launched Opus 5.5, it also cautioned that “benchmark margins have become a less reliable guide to real-world differences.”

Independent of Google, the Artificial Analysis Intelligence Index has tested Argon’s “High” setting. For testing, Artificial Analysis combines 10 evaluations covering math, science, coding, agentic tasks, and reasoning into a single composite score. Artificial Analysis says it runs all of the tests itself on standardized hardware, so it isn’t relying on each company’s self-reported numbers.

On that index, Gemini 4 Argon (High) scores 53, well above the median of 26 for comparable models. While still behind heavyweights like Claude and ChatGPT, Argon 4 represents a 23-point jump over Gemini 3.1 Pro Preview, Google’s previous top model.

At introductory pricing, Argon costs about $1.99 per Intelligence Index task, roughly 60 percent of GPT-6 Astra’s cost. Once the discount ends, that is expected to rise to $3.98, about 1.2 times Astra’s cost. By comparison, Opus 5.5 costs $5.98 per task.

Google Gemini 4 Argon: Safety and safeguards

Before today’s consumer rollout, Google released Argon in phases, starting with cyber defenders, while it participated in the U.S. government’s voluntary process for pre-release model access.

For everyone else, Google says the model is designed to refuse requests that could aid cyberattacks or chemical, biological, radiological, or nuclear attacks. Google says red teams tested those safeguards.

Google also says Argon is its most resistant model yet to indirect prompt injection. On Gray Swan’s benchmark — an Injection Prompt Test that measures how easily a model can be hijacked by malicious instructions hidden in content it reads — Google reports that attacks succeeded 0.7 percent of the time against Argon. The figures were 1.0 percent for both Opus 5.5 and Fable 5.1 and 8.5 percent for GPT-6 Astra. Lower is better on this test.

You can find the full Gemini Argon 4 model card on the Google website.

---

Disclosure: Ziff Davis, Mashable’s parent company, in April 2025 filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.

Source: https://mashable.com/tech/google-launches-gemini-4-argon