Google officially unveils Gemini 4 Argon

On September 30, 2026, Google officially unveiled Gemini 4 Argon — a frontier model with a 1M output token limit, compared to 64K in the previous generation, aimed at multi-step engineering coding, enterprise tasks, and cybersecurity.

Compiled by Sergey KostenchukPublished 2026-09-30Updated 2026-09-30

2026-09-30 Research OpenAI

🤖 Google officially unveils Gemini 4 Argon

On September 30, 2026, Google released a frontier model for multi-step tasks: engineering coding, enterprise work, and cybersecurity. The output token limit has been raised from 64K to 1M — a long task is solved in a single run.

🌍 The mechanics of agentic systems are changing: code migration and vulnerability patching proceed in a single trajectory. Inside Google, Argon agents migrated a Rust C/C++ codebase of up to 800K+ lines and freed up 300+ TiB of RAM. SOTA 77.9% on DeepSWE v1.1 and 68% on CWE-bench v1 are claimed.

👤 Access is currently limited to a closed test; expansion will begin with paid API clients and Google AI Ultra subscribers. The introductory price is $2 per 1M input tokens and $10 per 1M output tokens, then $4/$20. Benchmark figures are currently claimed by Google itself.

Source 1: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/

Source 2: https://9to5google.com/2026/09/30/gemini-4-argon-announcement/

Sources