[AINews] AI Engineer NYC: Gemini 4 Argon, Vals AI Benchmarks
Latent Space Latent.Space ● Covered by 12 sources
Gemini 4 Argon just went global, and the early benchmarks mostly back Google’s claims. At the same time, 20,000 people packed AIE NYC to argue over what these models are actually good for.
Based on reporting by Latent Space, Latent.Space — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Gemini 4 Argon is now live worldwide, and Google has spread it across a lot of surfaces at once: the paid Gemini API, the Gemini platform, Google AI Pro and Ultra in the app and Search’s AI Mode, plus Ultra in Antigravity. That’s a wider rollout than the earlier rumor mill suggested, and it puts the model directly in front of coders, enterprise users and security teams.
The pricing is sharp for now. Google is charging $2 per million input tokens and $10 per million output tokens on intro rates, with cached input cut by 95%. But that honeymoon ends after November 30, when Artificial Analysis says the rates rise to $4 and $20. That matters, because a lot of the current “cheap enough to use everywhere” math depends on the launch window.
Google is also talking up internal wins. It says Argon helped split core webpage handling out of a six-million-line Chromium folder, a job that took months and then got done in weeks. The company says the model drove 35% more agentic code submissions and cut rollbacks by nearly a third. Those are Google’s own deployment numbers, not a clean lab test, but they show where the company wants the story to go: less demo, more shipping.
The independent checks are strong enough to make that pitch sound plausible. Cursor puts Argon at 52.8% on CursorBench at default effort, roughly alongside Opus 5.5. Cognition reports 60.5% on FrontierCode 1.1, and says Devin users get a 75% discount through October 27. On the broader evaluation side, Artificial Analysis says Argon sits level with GPT-6 Astra on its Intelligence Index while costing about 60% as much per task under intro pricing.
The timing is messy in a very AI way. Argon went GA today, yet Latent Space says it already featured it in a title story two weeks ago because the benchmarks were out in public and mostly pointed the same direction. Meanwhile, about 20,000 people showed up at AIE NYC for talks from Anthropic, Vals AI and Coatue, which tells you where the attention is now: not whether the model exists, but what businesses are supposed to do with it once it does.
My take — AI-written commentary, not fact-checked reporting
This is the part of the AI cycle where launch day is almost secondary to benchmark day. Google can ship Argon globally and still be judged by a few public scores, because that’s how the market works now: numbers first, product second, reality sometime later. The annoying bit is that the numbers are useful only until the pricing changes, which is very on brand for the industry.
Read more about this at: Latent Space