Pokee-Isaac 28B: A 10M-Token Context Efficient Agentic Model

Abstract

Pokee-Isaac 28B is a non-decoder-only foundation model that reasons, plans, and uses tools over context windows of up to 10 million tokens. Capability at this context length has been assumed to require a model too large to deploy locally, and is today delivered almost exclusively from the cloud even when the supported context is much shorter. This leaves regulated, sovereign, and on-device settings, where data is not permitted to leave the boundary, with no path to it at all. Isaac runs on a single GPU — on one NVIDIA B200 it prefills at up to 137K tokens/s and decodes at 335 tokens/s — and is compact enough to serve on a client workstation. It holds retrieval fidelity across that full window, and matches or exceeds the strongest cost-optimized cloud systems on function calling, multi-turn interactive execution, tool orchestration, and terminal work. Deploying a long-context agentic model locally, inside the boundary where the data already sits, therefore becomes a real option.