New AI model wants to overturn competition in artificial intelligence - US Reflection with Greek director

A new, powerful artificial intelligence technology is preparing to change the balance in its global race AI, as startup Reflection, which is backed by Nvidia, prepares to launch its first model.

It aims to offer a cheaper and more adaptable alternative to companies such as Anthropic, OpenAI and Google, but also to compete with strong Chinese models.

Reflection AI is an American artificial intelligence startup, founded in 2024 by Misha Laskin and Ioannis Antonoglou, former researchers at Google DeepMind. Laskin is CEO and Antonoglou is Technical Director.

Its core philosophy is to develop powerful AI models with open parameters so that businesses, governments, and researchers can use, adapt, and install them in their own infrastructure, rather than relying on from closed systems such as ChatGPT or Claude.

The new model and the battle with China

According to sources quoted by Axios, Reflection's model is expected to be released soon. Initially, it will not be at the level of the most advanced American models, but it is expected to be competitive against the top Chinese open-ended models.

Although it will be one level below the most advanced American systems, it will have significant capabilities, enough that businesses can use it as a basis for developing their own, cheaper and specialized artificial intelligence systems.

In fact, the combination of a more economical model with high-quality data available to a company can, in specific applications, offer results comparable to those of much more expensive advanced models.

Open models are gaining ground

The so-called “open-weight” models, in which the system parameters are available so that they can be exploited and modified by third parties, have evolved into an important factor in the AI market. Unlike closed systems like ChatGPT and Claude, they can generally be downloaded for free and adapted to users' needs, that is.

The strongest models of this type today come largely from China and have gained a significant presence in Silicon Valley. On platforms that bring together many different models, their total share of usage has at times even exceeded 50%.

However, the picture is different in the business arena, where much more money is spent and AI is typically used through APIs and corporate billing. There open-weight models still account for a small percentage of total usage, according to industry executives and analysts.

At the same time, open models create a dilemma as they are more difficult for governments to monitor and regulate than large closed models, while their proponents note that the widespread availability and greater transparency can offer significant safety advantages.

The plan for 'AI factories'

Reflection's ultimate goal is even more ambitious, through the creation of the so-called “AI factory”, a product that will allow organizations to create their own, local AI ecosystems.

A company will, for example, be able to use its own proprietary data, Reflection models and its own computing power to create a highly customized and low-cost AI system.

The idea has already attracted the interest of organisations handling highly sensitive data. Hedge funds and trading firms, for example, are interested in developing their own systems using data they do not want to leak to external providers.

Reflection has already begun testing this model, announcing a partnership with South Korean conglomerate Shinsegae to create a “dominant AI factory,” Axios points out.

Agreements with Nebius and SpaceX

In view of the release of the model, Reflection has made contacts in Washington and elsewhere in recent weeks, presenting to stakeholders the capabilities of the system and explaining how the "AI factory" will work.

At the same time, the company is aggressively trying to secure the necessary computing power for its development. He recently signed major deals with Nebius and SpaceX in order to rent Nvidia chip-equipped AI servers.

Reflection CEO Misha Laskin, who has previously worked at Google, has warned that reaching the most advanced capabilities will take time.

“They're kind of like rocket boats. To build a large rocket-powered vessel, it takes time,” he told CNBC.