Google’s next big Gemini upgrade beats Claude and ChatGPT at their own game

Published Oct 1, 2026, 3:13 AM EDT I've been writing about Android since 2011, with a focus on device reviews, Samsung and Google Pixel hardware, and the latest happenings in the ecosystem. In my entire writing career, I've reviewed more than 75 Android phones. Carrying both a Samsung or Pixel flagship and an iPhone as a daily driver provides me with deep insight into how Android works and how it compares to iOS. I have been writing for Android Police since 2021, covering news, how-tos, and features. You can find my previous work on Neowin, AndroidBeat, Times of India, iPhoneHacks, MySmartPrice, and MakeUseOf. When not working, I tend to mindlessly scroll through X, play with new AI models, or go on long road trips. You can reach out to me on X or drop a mail at rajesh@androidpolice.com. Sign in to your Android Police account Google has been busy integrating Gemini across its products and services, but the AI itself falls well short of competitors like Claude and ChatGPT. For a company with Google’s resources and history in AI research, that’s a surprising position to be in. But the next big Gemini upgrade could change that, with Google teasing Gemini 4 Argon, which appears to outperform the best AI models from Claude and ChatGPT available today. Google says its latest frontier model, Gemini 4 Argon, promises sustained "deep reasoning across complex, long-horizon workflows" and delivers frontier-level performance in complex real-world tasks. Google is already using Argon in its internal workflows and says the model is fundamentally changing how the company works. A team of Argon agents applied memory optimizations across its data centers, freeing up 300 TiB of storage space. Google estimates the total savings could reach between 500 TiB and 1 PiB. Argon agents are also helping migrate Google’s C and C++ databases to Rush, working through tens of thousands of lines of code in core libraries. Google’s quantum computing researchers have used Argon to optimize spacetime resources, too. In one case, an agent beat the published baseline by 40% in just minutes. Based on the benchmarks Google shared across a range of tasks, including knowledge work, ML engineering, and agentic coding, Gemini 4 Argon beat Claude Opus 5.5, Fable 5.1, and ChatGPT-6 Astra, the latest models from Anthropic and OpenAI, respectively. It's also the leading model on the Vals Index, known for measuring AI models for their economic value across a wide range of sectors. To go along with Argon's improvements, Google is rolling out support for a 1M output token limit. That's a big jump from the 64K token limit of previous Gemini models. This is significantly higher than Anthropic's and OpenAI’s models, enabling Gemini to handle longer and more complex tasks and output in a single response. Gemini 4 Argon rolls out soon Unfortunately, you can't get your hands on Gemini 4 Argon just yet. Google is initially rolling out the model to trusted cyber defenders through its Fairwind Program. It wants to strengthen the model's safeguards before a public release. The company will refine its safeguards based on feedback from testers before expanding access. And even then, paid API customers and Google AI Ultra subscribers will be the first in line to try it out. For API users, Google will offer introductory pricing of $2 per million input tokens and $10 per million output tokens, with a 95% discount on cached tokens. While it's unclear how long the introductory pricing will last, after that, the prices will double to $4 and $20, respectively.