Alphabet shares gained about 2% in premarket trading Thursday after Google introduced Gemini 4 Argon, a new artificial intelligence model that the company says delivers stronger coding and cybersecurity capabilities at lower prices than competing models.
Google said Argon will first be made available to a select group of trusted cybersecurity researchers before its wider release.
Paid API customers and Google AI Ultra subscribers will receive access initially, with developers, businesses and consumers expected to follow.
Read: Man sues Google, blaming Gemini AI responsible for son’s suicide
The company is positioning Argon for demanding software development and cybersecurity tasks, among other professional applications. Google said the model can independently identify, verify and repair software vulnerabilities.
Argon recorded a 77.9% score on the DeepSWE v1.1 coding benchmark, which Google described as a record result.
Jefferies analysts said the model nevertheless trails leading systems on the more demanding FrontierSWE v2 evaluation.
The analysts also said Argon ranked first for coding on LLM Arena’s Text Arena. On Code Arena: Web Dev, the model placed eighth, 21 positions ahead of Google’s Gemini 3.8 Flash.
Cybersecurity performance is another area emphasized by Google and Jefferies and The brokerage said Argon matched Grok 4.7 for the highest score of 68% on CWE-bench v1, a test focused on fixing software vulnerabilities.
Jefferies also reported a 0.7% success rate for Argon in prompt-injection attacks where the comparable figure was 1% for Claude Opus 5.5 and Fable 5.1, according to the brokerage.
During the initial cybersecurity rollout, Google said Argon will be provided to trusted defenders and internal teams without some of its standard safeguards.
The company said it intends to add stronger protections before wider availability, including monitoring for possible misuse and intervening when needed to prevent harmful actions.
Cybersecurity company Wiz is using Argon through its Scan for Good program, Google said. The company also said it is already using the model internally for software development and data center optimization.
Argon’s pricing is another major part of the launch where Google is charging $2 per million input tokens and $10 per million output tokens and the cached input tokens are available at a 95% discount.
Jefferies said the launch pricing is below comparable offerings from Anthropic. Claude Fable 5.1 costs $10 per million input tokens and $50 per million output tokens, while Claude Opus 5.5 is priced at $4 and $20, respectively.
Google has also expanded Argon’s output limit to 1 million tokens from 64,000. Jefferies said Anthropic’s comparable models have a 128,000-token limit.
The company said Argon also performed strongly in evaluations involving automation, long-form video and professional work in fields including finance, legal services and tax.
Jefferies rates Alphabet as “buy” with a $445 price target whereas the brokerage said Alphabet shares had declined 16% from their May 18 peak amid investor concerns that the company could fall behind competitors while waiting for Gemini 4 Pro.
READ: Google employees must share data with AI tool to access health benefits: Report (October 9, 2025)
Jefferies cautioned that Argon’s early benchmark results will need to be validated through real-world use.
It also noted that the systems used to run and evaluate AI models are becoming increasingly important to overall performance.
Google said it plans to continue strengthening Argon’s safeguards and monitoring its use as the model moves toward broader availability.