Google unveils Gemini 4 Argon, the new artificial intelligence model challenging OpenAI and Anthropic

Google on Wednesday unveiled the new artificial intelligence model Gemini 4 Argon, as Reuters reports — the first model of the new Gemini 4 generation, arriving after months of delays and marking the tech giant’s attempt to catch up with rivals OpenAI and Anthropic in the AI race.

According to Google, Argon is the most powerful model DeepMind has ever built, larger than the previous “Pro” models and comparable to OpenAI’s GPT-6 Astra and Anthropic’s Opus in key coding and cybersecurity tests. The model supports deep reasoning on complex multi-step work, with an output limit of up to 1 million tokens, and leads in tests such as DeepSWE v1.1 for software engineering, as 9to5Google reports.

The starting API price — $2 per million input tokens and $10 per million output tokens — is one fifth of OpenAI’s prices for GPT-6 Astra, showing Google’s aggressive strategy to win over developers.

The rollout will be phased: initially the model will be given to trusted cybersecurity partners through the Fairwind program, and Google is taking part in the Trump administration’s voluntary process for early access to advanced models. The company has not yet given a date for the wider release. The announcement comes a day after chief executive Sundar Pichai signed the White House’s voluntary agreement on AI safety.

Alphabet shares rose in after-hours trading after the announcement, as investors welcomed Google’s return to the frontier of the most advanced models, nearly a year after the Gemini 3 generation.