Google has launched Gemini 4 Argon, a new AI model available to paid subscribers. The company reports improved performance in robotics and coding, with introductory API pricing set through November.
Google has released Gemini 4 Argon, its latest AI model, marking a significant update to its frontier AI lineup. The model is now available to specific subscriber tiers in the Gemini app and developer platforms, with introductory pricing for API access. Google states the model is designed for complex tasks in coding, science, and robotics.
Gemini 4 Argon is rolling out to users on Google AI Pro ($20/mo) and AI Ultra (starting at $99/mo) plans within the Gemini app and AI Mode in Search. Access to the model in Google Antigravity and AI Studio is restricted to AI Ultra subscribers. According to Ars Technica, Google has not clarified when or if Argon will be available to lower-tier or free users.
For developers using the API, Google has set an introductory price of $2 per million input tokens and $10 per million output tokens. Cached input tokens receive a 95 percent discount. This pricing is valid through November 30. After this period, the rates will increase to $4 per million input tokens and $20 per million output tokens.
Google reports that Argon achieves an 87% score on the Embodied Reasoning (ERQA) benchmark, which tests a model's ability to solve physical, spatial problems in 3D environments. The company states that Argon outperforms other frontier models on this specific benchmark. In a blog post, Google says the model marks a "step change in physical understanding, code generation, and visual-semantic reasoning."
The Verge reports that Google claims Argon competes with rival models such as GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 across benchmarks in coding, cybersecurity, science, finance, and law. Google also notes that Argon ranks #1 on the Vals Index, which measures performance across economic sectors.
Google has shared examples of how Argon is being used internally. According to Ars Technica, Argon helped optimize YouTube’s comment system by identifying redundant emoji rendering, which dropped server load by 45 percent. Engineers also used the model to find critical bugs in the drone physics simulator for Project Wing that had existed since 2018.
In a blog post, Google stated that Argon helped the Chrome team separate core webpage handling from a legacy folder within Chromium’s 40 million-line codebase. The company says this refactor, which involved hundreds of files, was completed in weeks rather than months. Additionally, Google reports that Argon improved the inference speed of its open Gemma models by 2.85x on an A100 GPU.
Google highlights Argon's capabilities in robotics, including controlling a Boston Dynamics Spot quadruped to dispose of trash and reconstructing simulation environments from video feeds. The company says Argon can generate new data in simulation to improve vision-language-action models, citing a boost in real-world robot success rates from 56% to 75% in specific tests.
In scientific research, Google reports that Argon helped design serine esterases, with laboratory testing confirming catalytic activity in five novel designs. The company also states that Argon assisted in discovering counterexamples to "Grad’s conjecture" in nuclear fusion research, alongside Professor Matt Landreman from the University of Maryland.
Sources
This article was drafted with AI assistance and checked against the sources above. Company claims are reported as claims. Cover image is AI-generated.
The day's checked news, by email
One short email a day with the top stories. Free, and you can unsubscribe in one click.
Or follow for the latest news:
AI & SITrending
Google DeepMind has released Gemini 4 Argon, a new frontier AI model available globally via API and subscriber platforms. The company reports improved coding efficiency and safety metrics, with introductory pricing set at $2 per million input tokens.
Anthropic researchers state small AI models may be months away from self-replicating on IoT devices. The company warns such systems would lack a central shutdown point.
