On Oct 13, 2026, Google announced that Gemini 4 Argon is now globally available to developers, enterprises and consumers. Google said the model is accessible via the API in AI Studio and the Gemini platform, to Google AI Ultra subscribers in Google Antigravity, and to Google AI Ultra and Pro subscribers in the Gemini app and AI Mode in Search.
Capabilities and benchmark claims
Google described Argon as a model designed for “deep work” across coding, finance, legal, cybersecurity and science. The company said Argon ranked #1 on the Vals Index, a benchmark that measures model performance across economic sectors such as finance, coding, legal and tax work.
Google reported concrete performance improvements from internal and partner use: teams saw agentic code submissions rise by more than 35% and code rollbacks fall by nearly a third. On a production open-source framework, Argon in Antigravity reportedly cut in-memory session prep time by 40% and sped up cold imports by 21%.
Researchers using Argon said it produced speedups in scientific toolchains: xTB was made up to 2.3x faster at predicting molecular structures and 1.8x faster at conformer search. Google also reported a 2.85x end-to-end inference speedup for Gemma 4 E2B on an A100 via vLLM, increasing throughput from ~37,900 to over 108,000 tokens per second with 100% token parity and no quality drop.
Use cases, research and robotics
Google described a range of real-world applications where Argon was used: large-scale codebase refactors inside Chromium’s 40 million-line codebase, automated optimization of scientific packages, multimodal 3D visualization and game prototyping using Blender and Three.js, and synthesis of large multimodal datasets into 3D reconstructions.
In scientific work, Google reported contributions spanning quantum chemistry, quantum computing, nuclear fusion and enzyme design. Examples cited include translating an internal AlphaQubit decoder to a hardware kernel that nearly doubled inference throughput, and Argon-produced implementations that reduced spacetime cost for certain quantum subroutines by more than 100x relative to a published baseline, according to Google.
On robotics benchmarks, Google said Argon achieved an 87% score on the Embodied Reasoning (ERQA) test. In an internal Video-to-Sim-to-Real test, Google reported that Apptronik’s Apollo humanoid replicated a cornhole skill and made seven consecutive shots after simulation transfer.
Security, safety and access
Google stated that Argon includes a multi-layer safety stack under its Frontier Safety Framework, and that external bio red teaming could only unlock less than 1% of tested bio sub-tasks. The company also described work with partners: Wiz used Argon in its “Scan for Good” initiative to identify and remediate vulnerabilities affecting government and healthcare infrastructure, according to Wiz and Google.
Google said the model launches at an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens discounted by 95%. The company also opened applications for 10,000 scientists to receive the highest usage tier in Google Antigravity free for six months, subject to eligibility and capacity.
Original source: Google AI