Article 79VZ3 Google continues model chemistry experiment with Neon

Google continues model chemistry experiment with Neon

by
Thomas Claburn
from www.theregister.com - Articles on (#79VZ3)
Story Image

Google periodically tables new AI models named after elements, and its latest offering has the noble name "Neon."

"Engineered for long-running agent workflows, Neon sustains performance through complex, tool-heavy tasks end-to-end - whether running standalone or acting as a fast, high-judgment Orchestrator directing parallel Hydrogen subagents in Google Antigravity," explained Koray Kavukcuoglu, SVP at Google DeepMind and chief AI architect at Google, in a blog post.

Neon emerged after Google last week emitted Hydrogen, after delivering Carbon two weeks earlier and wafting Argon out the door in late September.

Unlike OpenAI's celestial more intuitive model naming scheme - Astra, Sol, Luna, and Terra - which maps celestial position to capability, Google's chemical soup doesn't suggest a specific hierarchy.

To judge by price, Carbon or Argon would be Google's top tier model, based on non-discounted price that takes effect next month, each priced at $4/M input tokens, $20/M output.

The Chocolate Factory refers to Carbon as "Our most capable reasoning model" while Argon is "Our efficient Gemini 4 frontier reasoning model ... with faster responses and lower token spend per turn compared to Gemini 4 Carbon across complex coding, agentic, math, and science tasks."

Yet Neon, the latest element in the mix, surpasses both in Artificial Analysis' Intelligence Index, scoring 61 compared to Carbon at 60, Hydrogen at 54, and Argon at 50. At the same time, Neon costs less, at $3/M input tokens and $15/M output.

That favorable pricing is all the more surprising in light of Google's willingness, as of two days ago, to let users of its Gemini services opt out of generative AI model training without losing their chat histories.

Kavukcuoglu evidently recognizes the need to clarify its model rankings, an issue addressed in a blog post section titled "Where Neon sits in the Gemini 4 family."

"While Carbon tackles your hardest, long-horizon projects and Hydrogen handles fast everyday execution, Neon balances depth and efficiency," he explains. "Neon delivers frontier-level reasoning for complex, high-frequency workflows faster and more cost efficiently."

That may not really clarify which model is appropriate for which task and frankly Google is fine with that.

As noted, Google Antigravity comes with an Orchestrator that decides how to spend your money for you by swapping models on the fly. As the company describes it, a developer might start a task using Carbon, which would write a plan, and hand the task off to Hydrogen to benefit from its token thrift.

Antigravity does support model pinning, for those who know what they want. But Google, like other companies exploring model routing and orchestration, is fine with developers delegating decisions that affect billing to their trustworthy service provider. (R)

External Content
Source RSS or Atom Feed
Feed Location http://www.theregister.co.uk/headlines.atom
Feed Title www.theregister.com - Articles
Feed Link https://www.theregister.com/
Reply 0 comments