Gemini 4 Argon
Gemini 4 Argon, a new frontier model with a 1M-token output limit, scores 77.9% on DeepSWE v1.1; released first to vetted cyber defenders via Fairwind.
Phased rollout under the US voluntary pre-release access process; developers, enterprises and consumers (paid API, AI Ultra) come later. Google agents used it for C/C++ to Rust migrations and memory savings of 300+ TiB. Output limit rises from 64K to 1M tokens.
- Date
- Wednesday, 30 September 2026
- Lab
- Google DeepMind
- Kind
- model
- Access
- research preview
- Price
- Introductory $2 in / $10 out per 1M tokens, then $4 / $20 (announced 2026-09-30)
Figures
| Measure | Value | Measured by |
|---|---|---|
| DeepSWE v1.1 | 77.9% claimed state of the art on long-horizon software engineering | company |
| LVBench | 91.7% long video understanding, company-reported | company |
| AutomationBench (Zapier) | 51.3% ranked #1 per Google | company |
| CWE-bench v1 | 68% tied for first | company |
All figures are Google-reported; no public API at launch. The cyber build ships without cyber guardrails to trusted defenders. Google also monitors chain-of-thought for misalignment and urges industry to preserve reasoning transparency.
Sources
- deepmind.google/blog/gemini-4-argon-our-next-era-of-frontier-intelligence/
- blog.google/innovation-and-ai/technology/ai/google-ai-updates-september-2026/
This record was checked against its sources on 6 October 2026. How we check