Llama 4 对比 Mistral AI 与 MiniMax：4倍上下文差距 [2026]

文章对比了 Llama 4、Mistral AI 和 MiniMax 模型的上下文窗口大小。MiniMax 以 400 万 token 的上下文窗口领先，远超 Mistral AI 的 100 万 token 和 Llama 4 的 100 万 token。

Meta Mistral MiniMax

Moonshot AI unveils Kimi K3 model with 2.8 trillion parameters

Moonshot AI (Moon's Dark Side) has released its new Kimi K3 model with 2.8 trillion parameters, claiming it offers competitive performance while reducing deployment costs. The model uses a Mixture of Experts (MoE) architecture and reportedly achieves performance comparable to GPT-4o and DeepSeek-V3.

Moonshot AI