Google introduces new AI model

Google has released its latest artificial intelligence model, Gemini 4 Argon, aiming to provide a strong option for users of its cloud and workspace services. The model became available on October 13 and was tested against two major competitors: GPT-6 Astra from OpenAI and Claude Opus 5.5 from Anthropic. Benchmark tests showed that Gemini 4 Argon performs similarly to these models in many areas, with some advantages in specific tasks.

Performance in code execution

In tests measuring code execution, Gemini 4 Argon scored 57.4% on terminal code tasks. This is lower than Claude Opus 5.5, which scored 66.4%, and slightly higher than GPT-6 Astra, which scored 58.2%. These scores reflect the model's ability to handle programming tasks in a terminal environment, a common scenario for developers.

Strength in software engineering

Gemini 4 Argon excelled in software engineering tasks, scoring 77.9%. This outperformed both GPT-6 Astra (74.1%) and Claude Opus 5.5 (74.2%). The model's strong performance in this area suggests it may be particularly useful for developers working on complex coding projects.

Performance in scientific research

When it came to autonomous scientific research, OpenAI's GPT-6 Astra led with a score of 68.1%. Gemini 4 Argon scored 57.6%, and Claude Opus 5.5 scored 63.3%. This indicates that GPT-6 Astra is currently the top performer in this specific domain.

Mathematical reasoning abilities

Gemini 4 Argon scored 76.0% in mathematical reasoning, which is higher than both Claude Opus 5.5 (69.6%) and GPT-6 Astra (72.0%). This suggests the model is well-suited for tasks requiring strong numerical and logical skills.

Long video analysis capabilities

In long video analysis, Gemini 4 Argon scored 91.7%, outperforming both GPT-6 Astra (87.5%) and Claude Opus 5.5 (83.7%). This indicates that the model is particularly effective at understanding and analyzing extended video content.

Pricing structure

Gemini 4 Argon has a pricing model similar to Claude Opus 5.5, with input costs at $4 per million tokens, output at $20, and cache reading at $0.20. In comparison, GPT-6 Astra is significantly more expensive, with input costs at $10, output at $50, and cache reading at $1.00. Google is offering a discount until November 30, 2026, halving the input and output costs to $2 and $10, respectively.

Real-world application test

In a test to create a landing page for a fictional neo-cloud service, GPT-6 Astra produced the most complete and professional result. Gemini 4 Argon and Claude Opus 5.5 created more futuristic but less polished interfaces. This suggests that different models may be better suited for different types of creative tasks.

Language generation in French

For writing in French, Claude Opus 5.5 produced the most natural and fluent text. Gemini 4 Argon's output was more generic and repetitive, indicating that it may not be the best choice for tasks requiring high-quality multilingual content.

Video creation with Remotion

When creating a promotional video using the Remotion tool, both Gemini 4 Argon and Claude Opus 5.5 produced convincing results. Gemini 4 Argon's video had high-quality visuals but included excessive information, while Claude Opus 5.5's video was simpler and more effective. GPT-6 Astra's video was less polished and more animated than a typical motion design video.

Resource consumption

Gemini 4 Argon uses significantly more computational resources than its competitors, especially in agent workflows and video creation. This makes it less suitable for solo developers or entrepreneurs, despite its strong performance in most tasks. However, for users of Google Cloud and Workspace, the model remains a versatile and competitive option.