The Daily PulseTechnology
Google unveils Gemini 4 Argon AI model focused on coding and security
Reported by Faizan Khan (Staff Writer) · Indian Express - Technology ·
Reported by one source so far
Representative image · PexelsGoogle launched its newest frontier AI model, Gemini 4 Argon, on Wednesday, September 30, built to handle complex tasks that demand deep reasoning.
Capabilities and Performance
Designed for programming, logical thinking, and working with text and images, Argon excels at enterprise knowledge work like legal and financial tasks. Google stated that the model can independently find, check, and fix software errors, aiding cybersecurity defenders. During testing, the system found a security flaw in a hospital software package used worldwide that other advanced models missed. Argon set a record score of 77.9 per cent on the DeepSWE v1.1 benchmark for software engineering and topped the Vals Index for finance, coding, legal, and tax work. It also led on the AutomationBench leaderboard with 51.3 per cent and scored 91.7 per cent on LVBench for long video understanding. The model features an output limit of one million tokens, an increase from the previous 64,000 limit, allowing it to generate massive amounts of text in a single run.
Rollout and Safety
Because of its powerful cybersecurity skills, Google is restricting early access to trusted security partners in the Fairwind Programme to prevent misuse, with plans for a broader release to paid API users and Google AI Ultra subscribers later. CEO Sundar Pichai announced the model on social media, noting its strong performance in workflows and engineering. The model is priced at $2 per million input tokens and $10 per million output tokens, with a 95 per cent discount for cached input. Argon includes built-in protections against prompt injection, unauthorized sandbox escapes, and harmful requests related to cyber or biological threats.
Context
Advanced artificial intelligence models are undergoing strict scrutiny regarding safety as they gain autonomous capabilities to detect and fix software code. Tech companies often coordinate with government bodies on voluntary safety commitments to address risks of misuse while continuing to advance software capabilities.
ExplainerWhy this matters
What Happened
Google released Gemini 4 Argon, an advanced AI model capable of complex reasoning, software engineering, and cybersecurity tasks, while restricting its initial release to trusted partners due to safety concerns.
Why It Matters
As AI models become more autonomous and capable of finding software flaws or writing code, they also pose dual-use risks where bad actors or misaligned systems could exploit vulnerabilities. This has sparked intense debates among tech leaders and governments regarding how quickly advanced AI should be developed and what safeguards are necessary to prevent security breaches.
What Happens Next
Google plans to gather feedback from early testers in the Fairwind Programme, iterate on safety guardrails, and eventually make the model available to paid API customers, developers, enterprises, and consumers following participation in voluntary government safety assessments.
More from today
Representative image · Wikimedia CommonsScience & EnvironmentExplainerUnconfirmed — Reported by one source so farWorth knowing
Digestive Health and Environmental Rules Become Major Talking Points Ahead of US Midterms
Reported by Danish Ansari (Fact-Checker) · The Guardian - Environment ·
Representative image · PexelsHealth• UpdatedUnconfirmed — Reported by one source so farWorth knowing
Removing the ovaries (prophylactic oophorectomy) does not completely eliminate the risk of ovarian cancer because women can still develop primary peritoneal cancer that behaves like ovarian cancer.
Reported by Siddharth Deshmukh (Staff Writer) · Hindustan Times - Health ·
