Alphabet Rises as Google Debuts Low-Cost Gemini 4 Argon

Key Moments

Alphabet Shares Advance on New AI Launch

Alphabet Inc Class A (NASDAQ:GOOGL) moved about 2% higher in premarket trading on Thursday after Google rolled out Gemini 4 Argon, a new artificial intelligence model that the company is promoting as delivering stronger coding and cybersecurity capabilities than competing offerings at lower prices.

Rollout Strategy and Access Tiers

Google stated that Gemini 4 Argon will first be made available to a select group of trusted cybersecurity researchers before access is broadened. The company said paid API customers and Google AI Ultra subscribers will receive early access, followed by a wider release for developers, enterprises, and consumers.

Pricing Compared With Anthropic Models

Analysts at Jefferies flagged that Gemini 4 Argon is launching at $2 per million input tokens and $10 per million output tokens, which they described as significantly cheaper than offerings from Anthropic.

Google is also providing a 95% discount on cached input tokens, according to the same report.

Coding Benchmarks and Model Performance

Google reported that Gemini 4 Argon achieved a score of 77.9% on the DeepSWE v1.1 coding benchmark, which it described as a record result. Jefferies commented that the model falls short of top competitors on the more demanding FrontierSWE v2 benchmark, but pointed out that it ranked first for coding on LLM Arena’s Text Arena.

Jefferies also noted that Gemini 4 Argon placed eighth on Code Arena: Web Dev, representing an improvement of 21 positions relative to Gemini 3.8 Flash.

Cybersecurity Capabilities Under Scrutiny

Google indicated that Gemini 4 Argon can autonomously detect, validate, and remediate software vulnerabilities, framing cybersecurity as a key area of functionality that may attract investor attention.

Jefferies reported that the model tied Grok 4.7 for the highest score of 68% on CWE-bench v1, a benchmark that evaluates vulnerability remediation. Google also said that Argon showed a lower success rate for prompt-injection attacks, at 0.7%, compared with 1% for Claude Opus 5.5 and Fable 5.1, based on Jefferies’ account.

Early Cybersecurity Deployment and Safeguards

For the initial cybersecurity-focused deployment, Google said it will offer Gemini 4 Argon without some of its usual safeguards to trusted defenders and its own internal teams. The company added that cybersecurity firm Wiz is already using Argon through its Scan for Good program.

Looking ahead, Google said it intends to reinforce safety measures before a broader release. Planned steps include monitoring the model for misuse and intervening when necessary to stop harmful actions.

Broader Use Cases and Technical Limits

Google said Gemini 4 Argon demonstrated strong performance on tests related to automation, long-form video, and professional tasks in domains such as finance, legal services, and tax. The company has lifted the model’s output limit to 1 million tokens, up from 64,000, which Jefferies contrasted with 128,000 for Anthropic’s comparable models.

Google also said it is already deploying Argon internally for activities such as software development and data-center optimization. The article noted that these internal uses have not yet been independently verified as evidence of broader commercial uptake.

Analyst View on Alphabet and Validation Risks

Jefferies maintains a “buy” rating on Alphabet and set a price target of $445. The brokerage pointed out that the stock had dropped 16% from its May 18 high, amid concerns that Alphabet might lag competitors while the market awaited Gemini 4 Pro.

At the same time, Jefferies warned that real-world application will need to confirm the initial benchmark outcomes. The firm added that the systems used to operate and assess AI models are increasingly central to overall performance, rather than the models alone.