Gemini 4 Argon is Google's newest frontier AI model, announced via the company's official blog. The model provides an industry-leading 1 million token limit for both input and output, enabling deep, multi-step reasoning across long-horizon workflows. It is currently rolling out to trusted cyber defenders through the Fairwind Program, with a broader release planned after safety testing and guardrail refinement.

Performance evaluations show Gemini 4 Argon achieving a new state-of-the-art 77.9% on DeepSWE v1.1 for real-world software engineering tasks and leading scores on the Vals Index, Vals Finance Agent v2, Harvey's Legal Agent Benchmark, and AutomationBench (#1 at 51.3%). Internally, the model has optimized quantum algorithmic spacetime resources by 40% in minutes, identified memory savings exceeding 300 TiB across Google's data centers, and is migrating C/C++ codebases to Rust—including 800K+ lines for the Fuchsia Zircon kernel and replacing 32K lines of SIMD in libgav1 with safe Rust that yields a 2.7x faster video decoder.

Bar chart showing Gemini 4 Argon scoring 77.9% on DeepSWE v1.1 software engineering benchmark
Gemini 4 Argon achieves a new state-of-the-art 77.9% score on the DeepSWE v1.1 benchmark for real-world software engineering tasks.blog.google

The article cites Gemini 4 Argon's performance on DeepSWE v1.1 as a state-of-the-art result, indicating its strength in software engineering reasoning.

Chart displaying Gemini 4 Argon's leading score on the Vals Index benchmark
Gemini 4 Argon leads on the Vals Index benchmark, demonstrating strong performance in general reasoning tasks.blog.google

The article notes Gemini 4 Argon achieves leading scores on the Vals Index among other benchmarks.

Graph of Gemini 4 Argon's performance on Harvey's Legal Agent Benchmark
Gemini 4 Argon scores highly on Harvey's Legal Agent Benchmark, reflecting its capability in legal reasoning tasks.blog.google

The article mentions Harvey's Legal Agent Benchmark as one of the benchmarks where Gemini 4 Argon achieves leading scores.

Chart showing Gemini 4 Argon ranking first on AutomationBench with 51.3%
Gemini 4 Argon places first on AutomationBench with a score of 51.3%, indicating proficiency in automation reasoning.blog.google

The article states Gemini 4 Argon ranks #1 on AutomationBench at 51.3%, highlighting its automation capabilities.

In cybersecurity, Gemini 4 Argon can autonomously find, validate, and patch critical software vulnerabilities, tying for first place on CWE-bench v1 with a 68% score. Wiz has used the model through its Scan for Good initiative to uncover a critical vulnerability exposing personal information in healthcare software used by hospitals worldwide. Safety measures include the Frontier Safety Framework, monitoring internal activations to detect misuse, prompt injection robustness via adversarial training, misalignment mitigations that monitor chain-of-thought, and hardened sandboxed environments for secure testing.