Meta Lama 4 Benchmarking confusion: How well are the new Ai models?

There are two new Ai Models that the Slew of options names, thanks for new disclosures of meta. The company has created his lastory of general air models in Lamma 4 model and Lalma. 4 model

The competition between Meta and other Ai-firms is getting more intensely. Companies work for AI Models able to build and build in order to build more complex tasks and advanced ratification without requiring and running the computer’s computer power. It’s a Tricky sweet place to hit, and meta hoping his new models in front of the competitors like chatgtini.

Llama 4 benchmarking, explained

There treats some confusion about how new Lamama models stacles in comparison to the other models. There are a few ways to test chatbots, and one of the most part is by an unknown of ai filed platform called lmara, created by UC bellale Skylab. In the announcement of the Meta has claimed the company that his Maverick model has expressed, Hate-Chatagptt-4O.

But the mode that has been presented at the Linesa tests at all, is not the model that is available for using now. The model presented for tests becomes “llama-4-maverick-03-26-experimental.” In a footnote on a chart on the website (not the announcement), in little fortlasses in the final ballet point, meta that has embedd the model to LMentera. ”

Lmerenna a request on X / Twitters on Monday of Meta’s meta’s patreciation of the Mamerma. “In other words it is possible that Metaa is much better, more human-friendly model to meet their scorces. One way to do that it’s a model of data and testing processing, not ahead.

Meta’s VP of Generative Ai Ai Ahmad Alhle Tweeked that the company is trained on testing is trained “simply.” He said that the differences to see in the optics “is to stabilize due to the due to promptness.” As the publication, Meta’s Llama’s Maramick experimental (the original model is lined in the second place to LMENTS, with GMENTSA, with GTT-4O 4O- A PIPI 3.5. Google 3.5.

Meta did not direct an application for commentation.

Hits scout and maverick

There are two models in Lamama 4 relatives available now: Scout and Maverick. They are open weights models and multimodal, meaning that they have text, pictures and code. Open models such as META means that developers get some insights, when the models are built. Slowa 4 models are made up on open softly means that you see how the model improvements do in a variety of characterists get more weight than they’re getting more weight. Openai announces the previous month that it develops an open vicinity for the first time.

SCF is the small class model of the family, which designed to running on a single nvidia H100 GPU. The SCOUT has 10 million to corlassling fisters, and is it the 2 billion medicalities of Crafticks, the biggest of land a year the party will be given 178 experts. Mustomer Great with 12, -000 cames.

Some information about the rest of Llama 4: By the official model aman grace. We will probably learn more about these models in Llamaran, the first annual AI developer Conference the 29th April.

For more, look for what we know about a potential standalone app for Meta Ai and our review of the best AI Chatboats.