Meta is not on that list of “frontier AI” companies because it hasn’t kept up. As far as I know its most advanced model is Llama 4 and that’s not on the same level as GPT-5, Gemini, Grok, or Claude. Not only has it been left behind by the pivot to reasoning models; Meta’s special strength was supposed to be open source, but even there, Chinese models from Moonshot (Kimi K2) and DeepSeek (r2, v3) seem to be ahead. Of course Meta is now trying to get back in the game, but for now they have slipped out of contention.
The remaining question I have concerns the true strength of Chinese AI models, with respect to each other and their American rivals. You could turn my previous paragraph into a thesis about the state of the world: it’s the era of reasoning models, and at the helm are four closed-weight American models and two open-weight Chinese models. But what about Baidu’s Ernie, Alibaba’s Qwen, Zhipu’s ChatGLM? Should they be placed in the first tier as well?
I don’t think that they should. But December 2024 had Meta shock the world with first steps towards neuralese, even though they didn’t manage to scale it well. SOTA on neuralese is here, but nobody knows who will be the genius who creates a neuralese architecture which does scale.
You have missed Meta. Alas, it’s about as bad as xAI since it mostly produces AI companions...
Meta is not on that list of “frontier AI” companies because it hasn’t kept up. As far as I know its most advanced model is Llama 4 and that’s not on the same level as GPT-5, Gemini, Grok, or Claude. Not only has it been left behind by the pivot to reasoning models; Meta’s special strength was supposed to be open source, but even there, Chinese models from Moonshot (Kimi K2) and DeepSeek (r2, v3) seem to be ahead. Of course Meta is now trying to get back in the game, but for now they have slipped out of contention.
The remaining question I have concerns the true strength of Chinese AI models, with respect to each other and their American rivals. You could turn my previous paragraph into a thesis about the state of the world: it’s the era of reasoning models, and at the helm are four closed-weight American models and two open-weight Chinese models. But what about Baidu’s Ernie, Alibaba’s Qwen, Zhipu’s ChatGLM? Should they be placed in the first tier as well?
I don’t think that they should. But December 2024 had Meta shock the world with first steps towards neuralese, even though they didn’t manage to scale it well. SOTA on neuralese is here, but nobody knows who will be the genius who creates a neuralese architecture which does scale.