Pokee AI has released Pokee-Isaac 28B, a text base model with a context window of up to 10illion tokens, designed to operate within the customer's own infrastructure. According to MarkTechPost, the model weights have not been published; access is provided via license for VPC, on-premises deployment, or on-device usage.
In claimed results, the model scored 93,3% on RULER with a 10illion token context. In the comparison panel, all baseline solutions scored 0,0% beyond 2illion tokens. On BFCL v4, the model took first place with a score of 70,94, and on Terminal-Bench 2.1, it secured second place.
On a single B200 GPU, the prefill speed for the full context reached 137 200okens per second, while generation speed remained at approximately 335okens per second. The stated pricing is 0,15 per million input tokens and 1 per million output tokens.
editorial commentary
Why it matters
The likely consequence is increased interest in models that process large volumes of context within a customer-controlled environment. The next observable signal will be the publication of detailed testing methodologies, licensing terms, or independent verification of results. Significant uncertainty remains because currently only a synopsis from a single source is available, and the model weights have not been published.