no
>I wonder what technological aspect makes this possible.
one of its training datasets (prioritized somehow over the rest of them) contains a large number of examples emulating the thinking process within <think></think> tags before providing an output. the model then emulates it at runtime.
"Modern large language models (LLMs) started introducing a layer called “Mixture of Experts” (MoE) in their Transformer blocks to scale parameter count without linearly increasing compute. This is typically done through top-k (often k=2) “expert routing”, where each token is dispatched to two specialized feed-forward networks (experts) out of a large pool.
A naive GPU cluster implementation would be to place each expert on a separate device and have the router dispatch to the selected experts during inference. But this would have all the non-active experts idle on the expensive GPUs.
GShard, 2021 introduced the concept of sharding these feed-forward (FF) experts across multiple devices, so that each device"
[0] https://www.kernyan.com/hpc,/cuda/2025/02/26/Deepseek_V3_R1_...
models can communicate with one another via tool calling, sure, and there are hypothetical workflows where agents delegate tasks to other agents (with inference being done on different models), but that simply isn't the case here.