As AI models that run differently from large language models (LLMs) gain weight in the tech sector, related industries appear to be moving faster.

The industry said on Wednesday that more AI companies are seeking an edge with models that differ from LLMs, such as tabular foundation models (TFM) focused on structured data prediction, models that do not generate text and only make judgments and decisions, and code execution models (CEM) that operate inside running programs without human intervention and find more efficient ways to run code.

TFM supports companies in making predictions more easily based on structured tabular data organised in rows and columns, such as sales records, transaction histories and customer information.

IT market research firm Gartner picked large quantitative models (LQM) as one of the keywords to watch in the tech sector in 2027. LQM is a modular hybrid neural network architecture designed to process and analyse quantitative data, Gartner said. It is meant to support decision-making based on proactive simulation rather than reactive problem-solving.

The TFM market already has companies large and small. Global business applications company SAP acquired Pryor Labs, which specialises in TFM. Nvidia also recently released the open foundation model for tabular data prediction, Kumo Tabular, on open-source AI model sharing platform Hugging Face.

Kumo Tabular appears to be based on technology from Kumo AI, an enterprise predictive AI software startup Nvidia acquired in June. Kumo AI has developed AI models that answer predictive questions based on companies' structured data such as customer information and payment data. South Korean company Nums AI also unveiled its own LFM, Causilo, and moved in earnest to target the market.

Unlike LLMs that generate text, decision AI models that focus on probability-based judgment have emerged rapidly over about the past month in the AI sector.

After TypesafeAI, founded by former OpenAI employee Diogo Almeida (디오고 알메이다), unveiled jev in September, major companies such as OpenAI, Amazon Web Services (AWS) and Cloudflare also joined a race for decision models similar to jev.

For jev, if given a complex situation, a narrowly scoped question and a list of selectable answers, it provides decisions and probabilities instead of sentences. TypesafeAI said it would become a suitable AI model for AI agents.

More companies keep joining the decision model segment. Microsoft also unveiled Microsoft-Decision-1, an AI model specialised in fast decision-making.

Cloudflare released a preview of an AI model called Cleric, linked to its decision model Clef, that can extract information from documents or images quickly and cheaply. Cleric extracts names, dates and amounts from unorganised materials such as emails and photos of receipts. It also calculates or estimates values not in the materials and predicts numbers that could come next. It produces an answer at once without generating characters one by one like an LLM. Charges apply only to input data, and answers come within 1 second in most cases. If it cannot find an answer, it marks it as "none" rather than making up a plausible value. It also provides a confidence score with each answer.

Also drawing attention is CEM from DoubleAI, an Israeli startup backed by Nvidia. DoubleAI is emphasising CEM as a new AI category.

The Wall Street Journal reported that unlike LLMs, CEM operates inside running programs without human intervention and finds ways to run code more efficiently. DoubleAI's first product, Maxwell, helps address surging computing costs by boosting GPU and CPU performance.

Its revenue model is also interesting. DoubleAI takes a fee from part of the costs saved by customers. The company said one early customer cut computing costs by 67 percent. The AI company LTX doubled model inference speed with Maxwell, the WSJ reported.

DoubleAI was founded by CEO Amnon Shashua, who co-founded autonomous driving company Mobileye, and has received $220 million in investment so far from Lightspeed Venture Partners, Bessemer Venture Partners, Nvidia, Dell and others.