Gemini 4 Argon
14 hours ago
- Gemini 4 Argon is Google's new frontier model, initially rolled out to trusted cyber defenders via the Fairwind Program.
- It excels in deep reasoning for complex, long-horizon workflows, including software engineering, enterprise knowledge work (legal, finance), and cybersecurity defense.
- Pricing is $2 per million input tokens and $10 per million output tokens, with cached input tokens at 95% off.
- Internal use at Google shows improvements in quantum algorithmic optimization (40% better in minutes), memory efficiency (300 TiB+ saved), and large-scale codebase migrations (C/C++ to Rust).
- Output token limit expanded to 1M, enabling deep reasoning in a single trajectory.
- Achieves state-of-the-art performance on benchmarks: DeepSWE v1.1 (77.9%), Vals Index, LVBench (91.7%), AutomationBench (51.3%).
- Cybersecurity capabilities include autonomous vulnerability discovery and patching, demonstrated by finding critical healthcare software vulnerabilities missed by previous models.
- Safeguards include defenses against misuse, prompt injection attacks, misalignment monitoring, and hardened systems.
- Broad rollout will begin with paid API customers and Google AI Ultra subscribers after phased testing.