Gemini 4 Argon
Google's new flagship and the first Gemini 4 model: technically very capable, larger than Google's earlier Pro models, and literal-minded.
Quality
Modality
multimodal
Context
1M tokens
Access
closed
Fabian's Take
"I've only tested it a bit so far. It feels very competent technically, but it takes instructions too literally and isn't as creative as Opus 5.5, which has set the bar high. That's fine when you can say exactly what you want; for coming up with ideas I'd stay with Opus."
Gemini 4 Argon is the first model of Google’s Gemini 4 generation. Google announced it at the end of September, gave it to a small group of security partners first, and opened it to everyone on 13 October 2026. It’s not a renamed Pro model: according to Google it’s larger than its earlier Pro models, and it takes the top Gemini slot. The Gemini 3.5 Pro announced for the summer never shipped.
What it’s good at
Google positions Argon for long, multi-step work in coding, finance, legal research, cybersecurity and science, and says it follows style instructions better and hedges less than earlier Gemini models. The benchmarks Google published are its own, and they’re mixed: Argon leads on several, but in Google’s own table it trails the best competitor on two of the four coding benchmarks.
In the bit of testing I’ve done, the instruction-following shows. Argon does what you asked, and only what you asked. Where Opus fills a gap with a sensible idea of its own, Argon sticks to the letter of the prompt. Good for a precise technical task; less so when you want ideas.
Where to find it
Argon is in the Gemini app on the Google AI Pro and Ultra plans, in Google AI Studio, on the paid tier of the Gemini API, and in the Antigravity coding environment for AI Ultra subscribers. The introductory API price runs until 30 November 2026; from 1 December it doubles to $4 per million input tokens and $20 per million output tokens, which puts it at Opus 5.5’s price.
The Verdict
Best for: Precise technical work where you want instructions followed to the letter, like coding and analysis.
Pros
- Technically very strong; Google pitches it at coding, finance, legal, security and science work
- Follows instructions closely; Google says it also hedges less than earlier Gemini models
- 1M-token context and native image, audio, video and PDF input
- Introductory price of $2/$10 until the end of November
Cons
- Takes instructions too literally; you have to spell out what you'd leave implicit with Opus
- Less creative than Opus 5.5 in my tests
- The price doubles on 1 December 2026, to $4/$20
- Not on the free API tier
Specs
- Pricing $2/M input, $10/M output until 30 November 2026, then $4/$20 · batch half price · cached input 95% off · output includes thinking tokens
- Cost Tier moderate
- ⚡Speed Tier moderate
- License Proprietary
