Pokee AI launches 28B agent model, claims 10 million tokens on one RTX 4090

Pokee has not released the model's weights, detailed test configurations or enough architecture information to independently assess its long-context and throughput claims.

By · Published · Updated

Primary source: Pokee AI on X

Why it matters

A usable 10-million-token model on one GPU could cut agent inference costs and keep sensitive enterprise data on private hardware. Pokee's claims still need independent replication.

Pokee AI ships closed-source (for now) 28B model designed for 10-million-token workloads — Isaac is closed source, starts on RTX 4090-class hardware and targets private enterprise deployments, according to Pokee.

Zheqing (Bill) Zhu (@ZheqingZhu), founder and CEO of Pokee AI, launched Pokee-Isaac 28B on Tuesday, August 4th, with vendor-reported results suggesting the proprietary agent model can handle 10 million tokens of context on a single Nvidia RTX 4090-class GPU.…

When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.

Reader comments

Conversation for this story loads after sign-in.