Google Tops AI Smarts Ranking as OpenAI Slashes Prices

Google, OpenAI and Moonshot Set the Pace at the Top

• Google DeepMind announced Gemini 4 Carbon on Wednesday (Oct. 28), two weeks after its last top model, Gemini 4 Argon. Google calls Carbon its strongest model yet. Its biggest advance is in cybersecurity: finding software flaws, judging whether hackers could use them and writing fixes. Google’s Chrome team says Carbon performed more than 10 times better than another lab’s large frontier model at writing high-quality security fixes. Google says it now ranks first on the Artificial Analysis Intelligence Index, with a score of 59.
• OpenAI released GPT-6.1 Terra on Monday (Oct. 26), a mid-tier model at half the price of its predecessor. OpenAI says it handles computer tasks, such as filling forms and moving through software, more than three times faster than its flagship GPT-6 Astra and GPT-6.1 Sol models. On an internal OpenAI benchmark, the error rate fell from 53.6% for its predecessor, GPT-5.6 Terra, to 16.7%. It is available now to paid ChatGPT subscribers and to developers through OpenAI’s API.
• Moonshot AI launched Kimi K3.1 Pro on Oct. 23. The Beijing-based lab says its most capable model yet brings big gains in coding, 3D creation and financial research. Its demos include a linked financial model, an investment committee deck and a sourced industry report. The model can read roughly 1,500 pages at once, enough for a full deal file. Moonshot plans to make it free to download in early November.

Block and Rime Build Voice Models That Skip the Transcript

• Block introduced Dessa on Oct. 23, an early version of a speech model trained only on recorded human conversation, with no text at all. Dessa comes from a small lab inside the Square and Cash App parent that reports to Jack Dorsey. Block says Dessa listens and talks at the same time, laughs mid-sentence and hears sarcasm and frustration. Block frames it as a safety project: a voice AI that cannot tell when a caller is in distress just keeps going. It runs locally on a MacBook Pro M4 Max or better. Early access goes to a small group of AI safety researchers. The demo drew praise online and raised scam concerns.
• Rime launched Prism on Wednesday. The voice AI startup built it for enterprise call centers, where companies need to check what an agent will say before it says it. Prism drafts its reply as text first so compliance rules can screen every word, then speaks it while still listening to the caller. Rime says it answers in about half a second. On the telecom portion of τ-Voice, an Artificial Analysis benchmark built on simulated customer service calls, it scored 82.5 against 45.6 for OpenAI’s GPT-Live-1. Rime says companies can fine-tune it on their own call recordings, though self-serve tools are still in beta.

Nvidia and Liquid Push AI Onto Small Devices

• Nvidia released Nemotron 3.5 Bolt E4B on Oct. 23, a free model small enough to run on a PC without an internet connection. It is built for voice assistants and simple multistep tasks on local hardware.
• Liquid AI introduced LFM3 on Oct. 27, a family of six small free models that handle text, images, audio and quick decisions. Liquid says they run on phones and laptops and can be fine-tuned on a single graphics chip. Mercedes-Benz and Shopify tested them before release. Liquid pitches real-time fraud detection as one use case.

Cheaper models now handle the routine work. The frontier has moved to finding security holes before attackers do, and hearing a customer before the call goes wrong.

We’d love to be your preferred source for news.

Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!

Add as Preferred Source

For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.