Technology and Science · updated 2h ago · 2 min read
Google Unveils Gemini 4 Argon With 1M-Token Limit, Launching Soon for Subscribers
Gemini 4 Argon is Google's flagship frontier model. First in the Gemini 4 series, aims for coding and cybersecurity leadership.
Who gets Argon first, defenders vs partners.
8 of 10 outlets skipped it: initial pricing: $2 input tokens and $10 output tokens.
10outlets compared
Same story, two versions
tap a side to read it in full
The Straits Times
“Google said it was providing the model to select cybersecurity partners”Read the original ↗
TestingCatalog AI News
“Access is initially limited to trusted cyber defenders through Google’s Fairwind Program”Read the original ↗
The Straits Times stresses “partners”, while others foreground “trusted defenders” and Fairwind.
Gemini 4 Argon unveiled
Google announced Gemini 4 Argon as its “next era of frontier intelligence,” positioning the model for “deep reasoning across complex, long-horizon workflows.” Google said Gemini 4 Argon raises its output token limit to 1M tokens, up from 64K, to support longer and more complex use cases. Google said the model is built for frontier performance across coding, knowledge work, cybersecurity defense, and creative writing, and it is paired with expanded coding, reasoning, and multimodal capabilities.
Google said it trained Gemini 4 Argon to be “highly capable at cybersecurity defense,” and it described the model as rolling out “soon,” starting with Google AI Ultra subscribers and paid API customers. Google also said Argon will launch at an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off the input rate.
Limited access, internal use
Google limited early availability by making Gemini 4 Argon available first to a “set of trusted cyber defenders,” and Google said it is “actively engaged in the U.S. government’s voluntary process for pre-release model access.” Koray Kavukcuoglu said Google is “actively engaged in the U.S. government’s voluntary process for pre-release model access while we gradually expand access,” framing the phased approach as part of how the company expands access. Google said engineers inside the company are already using Argon extensively, and Ars Technica reported that “you aren’t allowed to use it yet.”
Google said Argon is already powering internal workflows, including teams using “fleet-wide telemetry data” to identify and apply memory optimizations across Google’s data centers. Google said Argon agents are migrating C/C++ codebases to Rust across Google, scaling from tens of thousands of lines in core libraries like re2 and libgav1 up to 800K+ lines for the Fuchsia OS Zircon kernel.
Benchmarks and safeguards
Google said Gemini 4 Argon scored 77.9% on DeepSWE v1.1, and it said the model ties for first place on CWE-bench v1 with a top score of 68%. Google said Argon ranks #1 on AutomationBench with a score of 51.3%, and it said the model is state of the art on LVBench with a score of 91.7%. Google said it is defending against misuse from cyber or chemical, biological, radiological, and nuclear (CBRN) attacks by improving “techniques to monitor the model’s internal activations to spot misuse.”
Google said it is also defending against prompt injection attacks with “the most resilient model yet against indirect prompt injections” and it is hardening sandboxed environments by isolating and sealing them before high-risk training or evaluations begin. Google said it is deploying misalignment mitigations that monitor Argon’s chain-of-thought and actions and stop execution when necessary, and it said it used a similar system to monitor training runs and send alerts to a dedicated incident response team.