{"id":666558,"date":"2025-05-02T14:48:47","date_gmt":"2025-05-02T11:48:47","guid":{"rendered":"https:\/\/en.buradabiliyorum.com\/ai2s-new-small-ai-model-outperforms-similarly-sized-models-from-google-meta\/"},"modified":"2025-05-02T14:48:47","modified_gmt":"2025-05-02T11:48:47","slug":"ai2s-new-small-ai-model-outperforms-similarly-sized-models-from-google-meta","status":"publish","type":"post","link":"https:\/\/buradabiliyorum.com\/en\/ai2s-new-small-ai-model-outperforms-similarly-sized-models-from-google-meta\/","title":{"rendered":"Ai2’s new small AI model outperforms similarly-sized models from Google, Meta"},"content":{"rendered":"
\n

\u2018Tis the week for small AI models, it seems. <\/p>\n

Nonprofit AI research institute Ai2 on Thursday released<\/a> Olmo 2 1B, a 1-billion-parameter model that Ai2 claims beats similarly-sized models from Google, Meta and Alibaba on several benchmarks. Parameters, sometimes referred to as weights, are the internal components of a model that guide its behavior.<\/p>\n

Olmo 2 1B is available under a permissive Apache 2.0 license on AI dev platform Hugging Face. Unlike most models, Olmo 2 1B can be replicated from scratch, as Ai2 has provided the code and data sets (Olmo-mix-1124<\/a> and Dolmino-mix-1124<\/a>) used to develop it.<\/p>\n

Small models might not be as capable as their behemoth counterparts, but importantly, they don\u2019t require beefy hardware to run. That makes them much more accessible for developers and hobbyists contending with the limitations of lower-end hardware and consumer machines.<\/p>\n

There\u2019s been a raft of small model launches over the past few days, from Microsoft\u2019s Phi 4 reasoning family to Qwen\u2019s 2.5 Omni 3B<\/a>. Most of these, including Olmo 2 1B, can easily run on a modern laptop or even a mobile device.<\/p>\n

Ai2 says Olmo 2 1B was trained on a data set of 4 trillion tokens from publicly available, AI-generated, and manually created sources. Tokens are the raw bits of data that models ingest and generate, with a million tokens equivalent to about 750,000 words.<\/p>\n

On a benchmark measuring arithmetic reasoning, GSM8K, Olmo 2 1B scores better than Google\u2019s Gemma 3 1B, Meta\u2019s Llama 3.2 1B, and Alibaba\u2019s Qwen 2.5 1.5B. Olmo 2 1B also eclipses the performance of those three models on TruthfulQA, a test for evaluating factual accuracy. <\/p>\n

\n
\n

Techcrunch event<\/p>\n

\n

\n\t\t\t\t\t\t\t\t\tBerkeley, CA<\/span>
\n\t\t\t\t\t\t\t\t\t\t\t\t\t|<\/span>
\n\t\t\t\t\t\t\t\t\t\t\t\t\tJune 5<\/span>\n\t\t\t\t\t\t\t<\/p>\n

\t\t\t\t\tBOOK NOW<\/span><\/p><\/div>\n<\/p><\/div>\n<\/div>\n

\n
\n
\n

This model was pretrained on 4T tokens of high-quality data, following the same standard pretraining into high-quality annealing of our 7, 13, & 32B models. We upload intermedia<\/a>te checkpoints from every 1000 steps in training.<\/p>\n

Access the base model: https:\/\/t.co\/xofyWJmo85<\/a> pic.twitter.com\/7uSJ6sYMdL<\/a><\/p>\n

\u2014 Ai2 (@allen_ai) May 1, 2025<\/a><\/p><\/blockquote>\n<\/div>\n<\/figure>\n

Ai2 has warned that Olmo 2 1B carries risks, however. Like all AI models, it can produce \u201cproblematic outputs,\u201d including harmful and \u201csensitive\u201d content, the organization said, as well as factually inaccurate statements. For these reasons, Ai2 recommends against deploying Olmo 2 1B in commercial settings.<\/p>\n<\/div>\n