Google Launches Gemini 4 Argon with One Million Token Context Window
Google has launched Gemini 4 Argon with a one-million-token context window, outperforming rivals on key benchmarks like DeepSWE v1.1 and LVBench while driving real-world robotics and cybersecurity breakthroughs.
After a prolonged period that felt like an eternity in the fast-moving AI race, Google has officially unveiled Gemini 4 Argon. The new model features a massive context window of one million tokens, a significant upgrade from the 64,000 tokens of previous generations, signaling Google's aggressive push to reclaim its market leadership.
Performance and Benchmarks
According to Google's official data, Gemini 4 Argon outpaces several prominent competitors across major industry benchmarks. The model achieved a record-breaking 77.9% on the DeepSWE v1.1 benchmark, alongside leading scores on the Vals Index. Furthermore, it secured the top spot on Zapier's AutomationBench with 51.3% and set a new benchmark high of 91.7% on LVBench for long-form video comprehension.
Despite these impressive achievements, competition remains fierce in specialized domains. OpenAI's GPT-6 Astra leads in the FrontierSWE v2 coding benchmark, while Anthropic's Claude Opus 5.5 outperforms rivals in Terminal-bench 4.0 and machine learning engineering tasks. Both models tied with a score of 68% on the CWE-bench v1 vulnerability remediation index.
Real-World Applications and Availability
Google highlights several practical applications of Gemini 4 Argon, including solving a historical mathematical problem, optimizing a quantum error decoder for hardware kernels, and assisting cybersecurity firm Wiz in discovering a critical vulnerability in hospital systems. Additionally, the model successfully analyzed a physical cornhole game from a video feed, generated simulation code, and enabled a humanoid robot to score seven consecutive successful throws in the real world.
Starting today, Gemini 4 Argon is accessible via the Gemini API, Antigravity, and Google AI Ultra and Pro subscriptions. Google has introduced an introductory pricing tier of $2 per million input tokens and $10 per million output tokens, featuring a 95% discount on cached tokens, before prices eventually double to standard rates.