On Wednesday local time, Alphabet (GOOGL.US) officially launched its flagship AI model Gemini 4 Argon to developers and enterprises worldwide. This is the first model in Google's Gemini 4 generation and currently sits at the top of its model lineup. Google CEO Sundar Pichai announced on X that the model is available globally starting immediately.
This move comes after months of launch delays and leadership changes at DeepMind, and is seen as a new round of direct competition against OpenAI and Anthropic. Google stated that Gemini 4 Argon excels at complex multi-step "deep work" in fields such as programming, finance, law, cybersecurity, and science.
In everyday chat scenarios, the model is designed to be more direct, less ambiguous, less repetitive, and less reliant on clichés; for longer tasks, it provides more frequent progress updates. Google also said that Argon has already taken on a significant amount of the company's internal engineering work, and this update embeds Google Antigravity's agentic coding interface directly into the macOS version of the Gemini app, allowing users to switch between chat and coding without changing applications.
It is reported that Google DeepMind Senior Vice President Koray Kavukcuoglu previewed Argon in a blog post on September 30, when the model was limited to trusted cyber defenders in Google's Fairwind program. With today's public release, Google AI Pro and Ultra subscribers can use Argon in the Gemini app, and AI Mode in Search can also access it; the Antigravity coding tool is limited to Ultra subscribers.
Google also announced that it will open the Fairwind Program to hundreds of new partners. The program provides a version of Argon without cyber guardrails to vetted cybersecurity defenders. In September, Google only offered the model to a select group of cybersecurity partners. Additionally, Google will provide 10,000 scientists with six months of free access to the highest usage tier of Antigravity.
On pricing, Google set an aggressive introductory API price for Argon: $2 per million input tokens, $10 per million output tokens, and a 95% discount on cached input tokens at $0.10 per million. According to the first source material, the relevant rates will double after November 30. By comparison, Anthropic's Claude Opus 5.5 charges $4 and $20 per million input and output tokens respectively, while OpenAI's GPT-6 Astra charges $10 and $50 respectively.
On subscription pricing, Google AI Pro is $19.99 per month. Google added a $100 per month Ultra plan at I/O 2026 and reduced the top-tier Ultra monthly fee from $250 to $200. The Gemini app also introduced several new features: a toggle in the main chat window allows switching between "chat" and "task" modes; in task mode, Gemini can continue working across Google apps and other services while the user is away from the computer, with Pro and Ultra subscribers having access to this capability. Gemini can now connect to more than 40 apps, including Instacart, OpenTable, Adobe, and Shopify, and Google positions it as capable of handling tasks such as ordering groceries, booking restaurant tables, and designing or building an online business.
Before Argon's release, Google's AI business had gone through a bumpy period. The company said in September it no longer planned to release Gemini 3.5 Pro, which Pichai initially said would launch in June. DeepMind also underwent restructuring after co-founder and CEO Demis Hassabis stepped back from the front line, and multiple Gemini model leads left the company.
The Argon launch comes during an autumn of密集 frontier AI model releases: Anthropic released Claude Opus 5.5, OpenAI released GPT-6 Astra, and Meta's "superintelligence" model Muse Wonder was released yesterday. On benchmark performance, Google's own evaluations show Argon leading on multiple metrics. In the Intelligence Index from independent organization Artificial Analysis, Argon with high reasoning mode enabled tied GPT-6 Astra at 53 points and was 1 point above GPT-6.1 Sol and the newly released Muse Wonder Max; however, the same index shows it trailing Claude Opus 5.5 and Claude Sonnet 5.5 by about 5 points. Google's comparison table includes 18 results, with Argon first in 14, tied with GPT-6 in cybersecurity, and behind in 5.
All the above data comes from Google, and AI companies typically also release evaluation results favorable to their own models. Google says Argon's largest lead is concentrated in knowledge work. It cites the Vals Index, which weights finance, programming, legal, and tax work according to each industry's share of U.S. GDP. On Zapier's AutomationBench, which tests end-to-end business tasks, Argon scored 51.3%, about 9 percentage points ahead of the next model.
Coding results are more mixed: Argon set a new high of 77.9% on Google's main test for long software engineering tasks, DeepSWE v1.1, but trailed GPT-6 Astra on FrontierSWE v2; on Terminal-Bench 4.0, Argon lagged all three competitors, with Opus 5.5 leading. Some of Google's competitor data also differs from Anthropic's reports. According to related reports, Anthropic listed GPT-6 Astra's score on Terminal-Bench 4.0 as 57.9%, while Google listed it as 58.2%; on the Terminal-Bench Science benchmark, Anthropic reported Opus 5.5 at 58.7% under default effort, while Google listed it as 63.3%. When releasing Opus 5.5, Anthropic warned that "benchmark gaps have become less reliable guides to real-world differences."
Artificial Analysis independently tested Argon's "High" setting. The index combines 10 evaluations spanning math, science, programming, agentic tasks, and reasoning into a single composite score, and says all tests were run on standardized hardware independently, without relying on companies' self-reported data. In that index, Gemini 4 Argon (High) scored 53, far above the median of 26 for similar models; although still behind heavyweight models such as Claude and ChatGPT, it jumped 23 points from Google's previous top model Gemini 3.1 Pro Preview. At introductory pricing, Argon costs about $1.99 per completed Intelligence Index task, roughly 60% of GPT-6 Astra's cost; after the discount ends, it is expected to rise to $3.98, about 1.2 times Astra's cost. By comparison, Opus 5.5 costs $5.98 per task.
On safety and protection, before the consumer launch, Google released Argon in stages, starting with cyber defenders, while also participating in the U.S. government's voluntary pre-release model access process. Google says that for ordinary users, the model is designed to refuse requests that could assist cyberattacks or chemical, biological, radiological, and nuclear attacks, and its related safeguards have been red-teamed. Google also says Argon is its most resistant model to indirect prompt injection to date. In Gray Swan's Injection Prompt Test benchmark, which measures how easily a model can be hijacked by malicious instructions hidden in content it reads, Google reported an attack success rate of 0.7% against Argon; Opus 5.5 and Fable 5.1 were both 1.0%, and GPT-6 Astra was 8.5%. In this test, lower values are better.
Comments