← All news·2026-09-30·3 min read

Google Introduces Gemini 4 Argon, Capable Of Producing Up To 1M Tokens

Google released a new model called Gemini 4 Argon that generates up to 1M tokens and achieved state of the art results on long development tasks.

aigoogle

On September 30th, Google announced their new Gemini 4 Argon model. This is a top-tier model designed for development, business applications, and cyber security. The new model can generate up to 1M tokens, which is an increase of almost 16x over the previous version (which supported 64,000 tokens). In their DeepSWE v1.1 benchmark, the new model achieved a new state of the art score of 77.9% for long development tasks. Currently, this model is only available to select cyber security experts through the Fairwind program. Pricing starts at $2/million tokens of input and $10/million tokens of output.

Not everyone can use it yet. Unfortunately, this model is still restricted to a small group of users. While the benchmark results are exciting, we won’t be able to evaluate this model ourselves until Google makes the API public. A 1M token output constraint is great for long agent chains because you won’t have to pause as often during large projects (e.g., refactoring a whole module at once). However, keep in mind that these are just theoretical results and not based on actual project data. Until I can try it out myself, I won’t know how well it stacks up against my current tools.

Source: blog.google

Free course

Stop reading about AI — start building with it

The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.

Start free →
EAEvgenii Arsentev

Author

Evgenii Arsentev

PhD · Founder & CEO, Dad4Kids