Gemini 4 Argon is officially released: what changed and how to use it
Google published Gemini 4 Argon on October 13, 2026. The stable API model is now documented, and this site’s live site catalog returns it as callable and priced.
The release is official
The Gemini API documentation names gemini-4-argon as the stable model code. Google’s model card was published on October 13, 2026, replacing the earlier uncertainty around availability with a concrete API and distribution record.
What Argon is designed for
Google positions Argon for complex software engineering, enterprise knowledge work, cybersecurity defense, deep visual understanding, science and mathematics. It accepts multimodal inputs and is designed for long, multi-step tasks.
What the official evaluations show
Google reports 77.9% on DeepSWE v1.1, 68.9% on Vals Index, 91.7% on LVBench and 84.2% on the 256k-to-1M GraphWalks setting. These are vendor-reported results; competing models lead other rows, so no single score settles every workload.
Limits worth knowing
The official model card warns about hallucinations, occasional slowness and timeouts, and gives March 2026 as the knowledge cutoff. Important outputs still need source checks, tests and human review.
Use it in this workspace
this site's live catalog returns gemini-4-argon-chat as available, callable and priced. Open the chat workspace, select Argon and check the server quote before sending.
Try the officially released model
The live selector exposes Argon only when this site's live catalog reports it as callable and priced.
Use Gemini 4 Argon