Mixtral 8x7B is a sparse mixture-of-experts large language model by Mistral AI, combining eight 7B expert networks with conditional routing for high efficiency. It targets performance competitive with much larger dense models (around Llama 2 70B level) while remaining relatively lightweight and fully open source.
Dense 70B open model from Meta; stronger single-token capacity but less efficient than sparse MoE for some workloads.
Proprietary OpenAI model with strong instruction-following and tooling ecosystem; accessed via API only.