Users were very excited about the arrival of Llama 4, the new AI model from Meta (Instagram, WhatsApp, Facebook, etc.). However, it seems we’ve suffered a kind of mirage. The results of its first benchmarks were quite powerful. So much so that it even surpassed ChatGPT and Gemini in several areas… but all this came with a fine print.
And certain reports indicate that the results could have been modified to show higher performance than the actual version. This means, in practice, that the version Meta initially made available to us is not the same as the final version. In this case, we’re talking about a fairly “experimental” version of Llama 4.
And to make matters worse, Meta has suspiciously delayed the development of Llama 4 for reasons only they really know. But one plus one equals two.
Flame 4: Questionable benchmarks and deceptive marketing
When Meta announced the Llama 4, they didn’t just want to sell us any old motorcycle; instead, they claimed it came with the equivalent of a Harley-Davidson under its arm. And this was enthusiastically received by users. Their benchmarks showed great performance results. But it seems the fuse on this firecracker has been very short.
Subsequent reports have pointed the finger at Meta itself, suggesting that those initial benchmarks were manipulated by the company itself. The goal was to show performance that was higher than it actually delivers. And obviously, this has sparked a firestorm within the community.
The Llama 4 Maverick, one of the most striking variants of this model, which was destined to dethrone the GPT-4o, has risen with authority in the LMArena rankings. The catch? The version of the Maverick that was presented was completely manipulated for the tests that were to be performed. Which, ultimately, means it’s not the same version that has reached end users.
And in addition to this controversy, now Llama 4 has been delayed by Meta due to “internal challenges.” But sources close to Meta have indicated that these are nothing more than new performance issues, specifically in the areas of math and reasoning. Therefore, Meta’s promises and the ambitious nature of its proposal have come to nothing.
Meta’s Response
For its part, Meta has denied all accusations of alleged benchmark manipulation. In fact, it reaffirms that Llama 4 will be a tremendous advance in the field of artificial intelligence. But the reality is that the community no longer believes in Meta this time around. The combination of these delays and doubts about the development of Llama 4 has only generated skepticism among users.
For now, I prefer to stick with GPT-4 or the latest Perplexity updates. This hasn’t been the only controversy Meta has had to face recently, as the company led by Mark Zuckerberg has been going from one step to the next.
