Google’s Carbon tops an AI ranking as rivals cut prices and push voice models

Google’s Gemini 4 Carbon has taken the top spot on the Artificial Analysis Intelligence Index, while competitors are cutting model prices and trying new approaches to voice AI. The mix points to a race measured not only in model rankings, but in security work, cost and how naturally systems handle speech.

Watch Desk analysis

What happened

Google DeepMind announced Carbon on 28 October, according to PYMNTS. The article says it scored 59 on the index. Google calls it the company’s strongest model yet, and says its Chrome team found it performed more than ten times better than another lab’s large frontier model at writing high-quality security fixes. That performance claim comes from Google, not an independent comparison described in the article.

The same report highlights a busy field: OpenAI halved GPT-6.1 Terra’s price to $1 per million input tokens and $5 per million output tokens; Block introduced its speech model Dessa for early access by safety researchers; and Rime launched Prism, which drafts call-centre replies as text so they can be screened before being spoken. Nvidia and Liquid AI also released small models intended to run locally on PCs or phones.

Why it matters

The useful signal is the spread of competition. Google is pitching security work, OpenAI is making a mid-tier model cheaper, and voice developers are tackling the awkward moment when an automated call needs to hear what someone means before answering. Smaller models that run locally offer another trade-off: less reliance on the cloud, with capabilities shaped by the device they can run on.

These are not directly comparable wins. The ranking, performance figures and product claims come from different tests and, in several cases, from the companies themselves. A leaderboard can tell us who scored highest on one index; it cannot settle which model is best for a particular job.

Our read

This is a more interesting AI race than a parade of ever-larger model names. Security fixes, lower prices and voice systems that can be checked before they speak all have practical consequences. But the numbers need their labels attached: a company benchmark is a claim to examine, not a universal performance guarantee. PYMNTS reports the announcements and figures discussed here.

What to watch

• Whether Carbon’s security-fix performance holds up in independent testing.
• How GPT-6.1 Terra performs on real workloads at its lower price.
• Whether Rime’s screening approach can meet call-centre compliance needs without slowing conversations.
• What tasks the new local models can handle on ordinary PCs and phones.

Discussion spark: Which would change your view of an AI model most: independently tested security work, a lower price, or voice that can be checked before it speaks?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.