News
General
45 views
Google unveils Gemini 4 Argon with coding and security capabilities
Oct 02, 2026
📍 Phliadelphia,PA, USA
### Alphabet Shares Rise After Google Unveils Gemini 4 Argon AI Model
Alphabet shares rose about 2% in premarket trading Thursday after Google introduced Gemini 4 Argon, a new artificial intelligence model designed to deliver stronger performance in software development and cybersecurity while offering lower pricing than several competing systems.
Google said Argon will initially be made available to a limited group of trusted cybersecurity researchers as the company evaluates its capabilities and safety in real-world environments. Paid API customers and Google AI Ultra subscribers will receive access first, followed by a broader rollout to developers, businesses and consumers.
Google is positioning Argon as a model for complex technical and professional workloads. The company said the system can independently identify software vulnerabilities, verify the findings and develop fixes without requiring continuous human intervention.
Argon achieved a 77.9% score on the DeepSWE v1.1 coding benchmark, which Google described as a record result. However, Jefferies analysts noted that the model still trails some leading systems on the more demanding FrontierSWE v2 evaluation.
The brokerage also said Argon ranked first for coding on LLM Arena’s Text Arena. On the Code Arena: Web Dev evaluation, Argon ranked eighth, although Jefferies noted that it was positioned 21 places ahead of Google’s Gemini 3.8 Flash.
Cybersecurity is another major focus of the launch. According to Jefferies, Argon achieved a 68% score on CWE-bench v1, matching Grok 4.7 for the highest result on the benchmark, which evaluates the ability of AI systems to identify and fix software vulnerabilities.
The brokerage also cited early results from prompt-injection testing. Argon recorded a 0.7% success rate in the attacks examined, compared with 1% for Claude Opus 5.5 and Fable 5.1, according to Jefferies.
During the initial cybersecurity deployment, Google said trusted security researchers and internal teams will receive access under conditions that remove some of the safeguards normally applied to the model.
Google said stronger protections will be introduced before broader availability. These measures are expected to include monitoring for potential misuse and interventions designed to prevent the system from carrying out harmful activities.
Cybersecurity company Wiz is already using Argon through its Scan for Good initiative, according to Google. The technology company also said its own teams are using the model for software development and data center optimization.
Pricing is another major part of Google's strategy for Argon. The company is charging $2 per million input tokens and $10 per million output tokens. Cached input tokens will be available at a 95% discount.
Jefferies said the launch pricing is below comparable offerings from Anthropic. The brokerage cited Fable 5.1 at $10 per million input tokens and $50 per million output tokens, while Claude Opus 5.5 is priced at $4 per million input tokens and $20 per million output tokens.
Google has also significantly expanded Argon's output capacity, increasing the limit to 1 million tokens from 64,000. Jefferies said comparable Anthropic models have a 128,000-token limit.
Beyond coding and cybersecurity, Google said Argon performed well in evaluations involving automation, long-form video and professional applications. These included use cases in areas such as finance, legal services and taxation.
The launch comes as investors continue to closely watch Google's position in the rapidly evolving generative AI market. Jefferies said Alphabet shares had fallen about 16% from their May 18 peak amid concerns that Google could lose ground to competitors while investors waited for Gemini 4 Pro.
Jefferies maintains a “buy” rating on Alphabet and a $445 price target, reflecting its assessment of the company's broader business and AI prospects.
At the same time, the brokerage cautioned that Argon's early benchmark results will need to be tested through broader real-world use before its performance can be fully assessed.
Jefferies also highlighted the growing importance of the infrastructure used to train, run and evaluate AI models, suggesting that model performance alone may not determine how competitive these systems become.
Google said it will continue improving Argon's safety measures and monitoring its use as the model moves from its limited cybersecurity rollout toward wider availability.
Alphabet shares rose about 2% in premarket trading Thursday after Google introduced Gemini 4 Argon, a new artificial intelligence model designed to deliver stronger performance in software development and cybersecurity while offering lower pricing than several competing systems.
Google said Argon will initially be made available to a limited group of trusted cybersecurity researchers as the company evaluates its capabilities and safety in real-world environments. Paid API customers and Google AI Ultra subscribers will receive access first, followed by a broader rollout to developers, businesses and consumers.
Google is positioning Argon as a model for complex technical and professional workloads. The company said the system can independently identify software vulnerabilities, verify the findings and develop fixes without requiring continuous human intervention.
Argon achieved a 77.9% score on the DeepSWE v1.1 coding benchmark, which Google described as a record result. However, Jefferies analysts noted that the model still trails some leading systems on the more demanding FrontierSWE v2 evaluation.
The brokerage also said Argon ranked first for coding on LLM Arena’s Text Arena. On the Code Arena: Web Dev evaluation, Argon ranked eighth, although Jefferies noted that it was positioned 21 places ahead of Google’s Gemini 3.8 Flash.
Cybersecurity is another major focus of the launch. According to Jefferies, Argon achieved a 68% score on CWE-bench v1, matching Grok 4.7 for the highest result on the benchmark, which evaluates the ability of AI systems to identify and fix software vulnerabilities.
The brokerage also cited early results from prompt-injection testing. Argon recorded a 0.7% success rate in the attacks examined, compared with 1% for Claude Opus 5.5 and Fable 5.1, according to Jefferies.
During the initial cybersecurity deployment, Google said trusted security researchers and internal teams will receive access under conditions that remove some of the safeguards normally applied to the model.
Google said stronger protections will be introduced before broader availability. These measures are expected to include monitoring for potential misuse and interventions designed to prevent the system from carrying out harmful activities.
Cybersecurity company Wiz is already using Argon through its Scan for Good initiative, according to Google. The technology company also said its own teams are using the model for software development and data center optimization.
Pricing is another major part of Google's strategy for Argon. The company is charging $2 per million input tokens and $10 per million output tokens. Cached input tokens will be available at a 95% discount.
Jefferies said the launch pricing is below comparable offerings from Anthropic. The brokerage cited Fable 5.1 at $10 per million input tokens and $50 per million output tokens, while Claude Opus 5.5 is priced at $4 per million input tokens and $20 per million output tokens.
Google has also significantly expanded Argon's output capacity, increasing the limit to 1 million tokens from 64,000. Jefferies said comparable Anthropic models have a 128,000-token limit.
Beyond coding and cybersecurity, Google said Argon performed well in evaluations involving automation, long-form video and professional applications. These included use cases in areas such as finance, legal services and taxation.
The launch comes as investors continue to closely watch Google's position in the rapidly evolving generative AI market. Jefferies said Alphabet shares had fallen about 16% from their May 18 peak amid concerns that Google could lose ground to competitors while investors waited for Gemini 4 Pro.
Jefferies maintains a “buy” rating on Alphabet and a $445 price target, reflecting its assessment of the company's broader business and AI prospects.
At the same time, the brokerage cautioned that Argon's early benchmark results will need to be tested through broader real-world use before its performance can be fully assessed.
Jefferies also highlighted the growing importance of the infrastructure used to train, run and evaluate AI models, suggesting that model performance alone may not determine how competitive these systems become.
Google said it will continue improving Argon's safety measures and monitoring its use as the model moves from its limited cybersecurity rollout toward wider availability.
Tags
news
Comments (0)
Login to post comments
No comments yet
Be the first to share your thoughts about this post.