Despite the fierce AI arms race, we face a multimodal future

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. learn more

Every week – sometimes every day – a new one State-of-the-art artificial intelligence model Born into the world. As we enter 2025, the pace of new model releases is dizzying, if not exhausting. The curve of the roller coaster continues to grow exponentially, with fatigue and surprise becoming constant companions. Each version highlights why this Certain models outperform all others, filling our feeds with endless collections of benchmarks and bar charts as we scramble to keep up.

Since 2020, the number of large-scale basic models published every year has shown explosive growth
Charlie Giattino, Edouard Mathieu, Veronika Samborska and Max Roser (2023) – “Artificial Intelligence” Published online at OurWorldinData.org.

Eighteen months ago, the vast majority of developers and enterprises were using Single AI model. Today, the opposite is true. Few sizable enterprises limit themselves to the capabilities of a single model. Companies are wary of vendor lock-in, especially for a technology that has quickly become a core part of long-term corporate strategy and short-term bottom-line revenue. It’s increasingly risky for teams to bet everything on a single large language model (LLM).

But despite this fragmentation, many model providers remain convinced that AI will be a winner-take-all market. They claim that the expertise and computation required to train state-of-the-art models is scarce, defensible, and self-reinforcing. From their perspective, the hype bubble Construct artificial intelligence models It will eventually collapse, leaving behind a giant artificial general intelligence (AGI) model that will be used for everything. Exclusive ownership of this model means becoming the most powerful company in the world. The size of this award has sparked an arms race for more and more GPUs, with the number of training parameters increasing by a new zero every few months.

Deep Thought, Holistic AGI in The Hitchhiker’s Guide to the Universe
BBC, The Hitchhiker’s Guide to the Galaxy, TV series (1981). Retrieve static images for review purposes.

We believe this view is wrong. No single model will rule the universe next year or in the next decade. Instead, the future of AI will be multi-model.

Language models are fuzzy commodities

this Oxford Economic Dictionary Commodities are defined as “standardized goods that are bought and sold on a large scale and whose units are interchangeable.” Language models are commodities in two important senses:

The models themselves become more interchangeable across a wider range of tasks;
The research expertise required to produce these models is becoming more distributed and accessible, with cutting-edge labs barely able to outdo one another and independent researchers in the open source community following suit.

*An item that describes the product (Source: Not Diamond)*

But while language models are becoming commoditized, progress is uneven. Any model (from GPT-4 all the way up to Mistral Small) is well suited to handle a large number of core functions. At the same time, as we move toward edges and edge cases, we see increasing differentiation, with some model providers explicitly focusing on code generation, inference, retrieval-augmented generation (RAG), or mathematics. This resulted in endless hand-wringing, Reddit scouring, evaluation, and fine-tuning to find the right model for each job.

*AI models are commoditizing around core functionality and specializing at the edges. Credit: Not a Diamond*

So while language models are commodities, they are more accurately described as Fuzzy goods. For many use cases, AI models are nearly interchangeable, with metrics like price and latency determining which model to use. But at the edge of capability, the opposite will happen: models will continue to specialize and become increasingly differentiated. For example, Deepseek-V2.5 Although it is a fraction of the size and 50 times cheaper than GPT-4o, it is more powerful than GPT-4o for C# coding.

These two dynamics – commoditization and specialization – undermine the argument that a single model is best suited to handle all possible use cases. Instead, they point to the gradual fragmentation of the AI landscape.

There is an apt analogy for the market dynamics of language models: the human brain. The structure of our brains has remained unchanged for hundreds of thousands of years, and our brains are far more similar than different. For the vast majority of our time on earth, most people have learned the same things and have similar abilities.

But then things changed. We develop the ability to communicate with language—first orally, then in writing. Communication protocols promoted the development of the Internet, and as humans began to connect to the Internet, we also began to become more and more specialized. We are freed from the burden of being generalists in all areas, of being islands of self-sufficiency. Paradoxically, the collective wealth of specialization also means that the average person today is a stronger generalist than any of our ancestors.

Over a wide enough input space, the universe always tends to specialize. This is true from molecular chemistry to biology to human society. If there is enough diversity, a decentralized system will always have higher computational efficiency than a monolithic system. We believe the same is true for artificial intelligence. The more we can leverage the strengths of multiple models, rather than relying on just one, the more those models can be specialized, thereby expanding the frontier of capabilities.

*Multi-model systems enable greater specialization, functionality, and efficiency. Source: Not Diamond*

An increasingly important pattern that takes advantage of different models is routing—dynamically sending queries to the most suitable model while taking advantage of cheaper, faster models without degrading quality. Routing allows us to take advantage of all the benefits of specialization—higher accuracy, lower cost and latency—without giving up any of the robustness to generalization.

A simple demonstration of routing capabilities can be seen in the fact that most of the top models in the world are routers themselves: they are built using Expert blend An architecture that routes each next-generation token to dozens of expert sub-models. If LLM is indeed exponentially growing the fuzzy commodity, then routing must become an essential part of every AI stack.

There is an argument that when LL.M.s reach human intelligence, they will level off – when our capabilities are fully saturated, we will consolidate around a common model, just like we consolidated around AWS or the iPhone. None of these platforms (or their competitors) have improved their capabilities 10x in the past few years, so we might as well adapt to their ecosystems. However, we believe that artificial intelligence will not stop at human-level intelligence; it will far exceed any limit we can imagine. As it does so, it will become increasingly decentralized and specialized, just like any other natural system.

We cannot overstate that AI model fragmentation is a good thing. Decentralized markets are efficient markets: they empower buyers, maximize innovation and minimize costs. To the extent that we can leverage a network of smaller, more specialized models rather than sending everything through the guts of a single giant model, we will move toward a safer, more explainable, and more controllable AI future.

The greatest inventions have no owners. Ben Franklin’s heirs had no electricity. Turing’s legacy does not own all computers. Artificial intelligence is undoubtedly one of humanity’s greatest inventions; we believe its future will and should be multimodal.

Zack Kass is Open artificial intelligence.

Tomás Hernando Kofman is Not a diamond.

data decision maker

Welcome to the VentureBeat community!

DataDecisionMakers is a place where experts, including technologists working in data, can share data-related insights and innovations.

If you want to stay up to date on cutting-edge thinking and the latest news, best practices and the future of data and data technologies, join us at DataDecisionMakers.

you might even consider Contribute an article Your own!