Google releases Gemini 4 Argon: How to strive it, pricing, benchmarks

Google introduced the general public launch of Gemini 4 Argon after an earlier restricted launch to pick out testing companions. It’s the primary mannequin within the firm’s Gemini 4 era and now sits on the high of Google’s lineup.

Google CEO Sundar Pichai introduced the launch on X, the place he mentioned the mannequin can be obtainable globally beginning at the moment.

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

Google first introduced Argon on Sept. 30 in a weblog submit by Google DeepMind SVP Koray Kavukcuoglu. At that time, it was restricted to trusted cyber defenders in Google’s Fairwind Program. With at the moment’s launch, Google AI Pro and Ultra subscribers can use Argon within the Gemini app.

Google says Argon is constructed for lengthy, advanced work in software program engineering, enterprise information work like authorized and finance, and cybersecurity protection. For on a regular basis chat, Google says the mannequin will get to the purpose with much less hedging, repeats itself much less, and avoids clichés. On longer duties, it offers extra frequent progress updates.

Argon 4 arrives throughout a busy fall season for frontier AI fashions. Google repeatedly missed its personal deadlines for the launch of Gemini 3.5 Pro, and in the meantime, Anthropic launched Claude Opus 5.5, and OpenAI launched GPT-6 Astra. Meta’s “super intelligence” Muse Wonder was simply launched yesterday.

As anticipated, Google’s personal analysis reveals Argon testing forward of its rivals on many benchmarks. On the impartial Artificial Analysis Intelligence Index, Argon ties GPT-6 Astra at 53 with excessive reasoning enabled, one level above GPT-6.1 Sol and the brand new Muse Wonder Max. The similar index locations it about 5 factors behind Claude Opus 5.5 and Claude Sonnet 5.5.

Coding is the clearest focus of the discharge. Google says Argon already handles a lot of its inner engineering work, and at the moment’s replace brings Google Antigravity’s agentic coding interface straight into the Gemini app on macOS.

Want to be taught extra about getting the most effective out of your tech? Sign up for Mashable’s Top Stories and Deals newsletters at the moment.

Google Gemini Argon 4: How to strive it

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

Google set an introductory API worth for Argon:

• $2 per million enter tokens

• $10 per million output tokens

• Cached enter tokens at 95 % off, or $0.10 per million

That is lower than each principal rivals cost. Anthropic’s Claude Opus 5.5 prices $4 and $20 per million enter and output tokens, whereas GPT-6 Astra prices $10 and $50 per million tokens. Today’s announcement focuses on the Gemini app and doesn’t say whether or not API entry has widened past Google’s earlier rollout plans.

The rollout comes with a number of different modifications to the Gemini app. First, Antigravity’s agentic coding interface now lives contained in the Gemini app on macOS, so customers can transfer between chatting and coding with out switching apps. Second, a brand new toggle in the principle chat window switches Gemini from chat to duties. In duties mode, Gemini can preserve working throughout your Google apps and different companies whilst you’re away out of your pc. Pro and Ultra subscribers get entry.

Additionally, Gemini now connects to greater than 40 apps, together with Instacart, OpenTable, Adobe, and Shopify. Google pitches this for jobs like ordering groceries, reserving restaurant tables, and designing or constructing a web based enterprise.

Google Gemini Argon 4: Benchmark efficiency

Google’s comparability desk has 18 outcomes. Argon has the highest rating on 14, ties with GPT-6 in cybersecurity, and trails on 5. All figures on this part come from Google. Typically, AI corporations launch outcomes that present their very own mannequin scoring above its rivals.

Credit: Google / Mashable screenshot

Argon’s greatest leads are in information work. Google says it tops the Vals Index, which weights finance, coding, authorized, and tax work by every sector’s share of U.S. GDP. On AutomationBench, Zapier’s take a look at of end-to-end enterprise duties, Argon scored 51.3 %. That is about 9 factors forward of the subsequent mannequin.

The coding outcomes are blended. Argon set a brand new excessive on DeepSWE v1.1, Google’s principal take a look at of lengthy software program engineering duties, at 77.9 %. 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 rivals don’t match what Anthropic reported. Mashable reported that Anthropic put GPT-6 Astra at 57.9 % on Terminal-Bench 4.0, whereas Google lists 58.2 %. On the Terminal-Bench Science benchmark, Anthropic reported 58.7 % for Opus 5.5 (default effort), whereas Google lists 63.3 %. When Anthropic launched Opus 5.5, it additionally cautioned that “benchmark margins have become a less reliable guide to real-world differences.”

Independent of Google, the Artificial Analysis Intelligence Index has examined Argon’s “High” setting. For testing, Artificial Analysis combines 10 evaluations overlaying math, science, coding, agentic duties, and reasoning right into a single composite rating. Artificial Analysis says it runs the entire assessments itself on standardized {hardware}, so it isn’t counting on every firm’s self-reported numbers.

On that index, Gemini 4 Argon (High) scores 53, nicely above the median of 26 for comparable fashions. While nonetheless behind heavyweights like Claude and ChatGPT, Argon 4 represents a 23-point soar over Gemini 3.1 Pro Preview, Google’s earlier high mannequin.

Argon ranks only a hair above Muse Wonder however beneath Anthropic and OpenAI’s newest.
Credit: Artificial Analysis

At introductory pricing, Argon prices about $1.99 per Intelligence Index activity, roughly 60 % of GPT-6 Astra’s value. Once the low cost ends, that’s anticipated to rise to $3.98, about 1.2 occasions Astra’s value. By comparability, Opus 5.5 prices $5.98 per activity.

Google Gemini 4 Argon: Safety and safeguards

Before at the moment’s shopper rollout, Google launched Argon in phases, beginning with cyber defenders, whereas it participated within the U.S. authorities’s voluntary course of for pre-release mannequin entry.

For everybody else, Google says the mannequin is designed to refuse requests that would help cyberattacks or chemical, organic, radiological, or nuclear assaults. Google says crimson groups examined these safeguards.

Google additionally says Argon is its most resistant mannequin but to oblique immediate injection. On Gray Swan’s benchmark — an Injection Prompt Test that measures how simply a mannequin could be hijacked by malicious directions hidden in content material it reads — Google experiences that assaults succeeded 0.7 % of the time in opposition to Argon. The figures had been 1.0 % for each Opus 5.5 and Fable 5.1 and eight.5 % for GPT-6 Astra. Lower is best on this take a look at.

You can discover the total Gemini Argon 4 mannequin card on the Google web site.

Disclosure: Ziff Davis, Mashable’s dad or mum firm, in April 2025 filed a lawsuit in opposition to OpenAI, alleging it infringed Ziff Davis copyrights in coaching and working its AI techniques.