The new small AI model of AI2 is superior to models of similar sizes from Google, Meta

“This is the week for small AI models, it looks.

On Thursday, AI2, the Non -profit Research Institute of AI, released Olmo 2 1b, a 1 billion parameter model for which AI2 claims, defeats models of a similar size from Google, Meta and Alibaba in several indicators. The parameters, sometimes called weights, are the internal components of a model that guides its behavior.

Olmo 2 1b is available under Apache 2.0 license to the hug of the AI ​​DEV platform. Unlike most models, Olmo 2 1B can be replicated from scratch; AI2 has provided codes and data sets (OLMO-MIX-1124., Dolmino-Mix-1124) Used to develop it.

Small models may not be as capable as their behemoth counterparts, but the important thing is that they do not require Beefy hardware to work. This makes them much more accessible to developers and lovers who are struggling with lower-class machines and users.

In the last few days there have been a raft of small model launches, from Microsoft Phi 4 Reflection Family yes 2,5 oman on qwen 3bS Most of them – and Olmo 2 1b – can easily operate on a modern laptop or even a mobile device.

AI2 says Olmo 2 1b has been trained in a set of data of 4 trillion tooth from publicly available, AI generated and hand -created sources. Tokens are the raw bits of data models that accept and generate – 1 million tokens are equivalent to about 750,000 words.

Of arithmetic measurement indicator, GSM8K, OLMO 2 1B ratings better than GEMMA 3 1B of Google, Llama of Meta 3.2 1B and QWEN 2.5 1.5B of Alibaba. Olmo 2 1b also darkens the operation of these three Amestfulqa models, a test to evaluate actual accuracy.

However, AI2 warns that OLMO 2 1B carries risks. Like all AI models, it can produce “problematic results”, including harmful and “sensitive” content, says the organization, as well as actually inaccurate statements. For these reasons, AI2 recommends that Olmo 2 1b be deployed in commercial settings.