AI Playground 3.0.3 caps context at 8K

No public evidence suggests a 288V-only governor; this looks like conservative runtime guardrails in the OpenVINO/NPU path rather than a model ceiling. Intel’s 3.0.3 notes explicitly list Core Ultra Series 2 (V) / Lunar Lake support. Intel added context-size support for OpenVINO on NPU devices in a recent release cycle, which suggests the effective limit is backend-driven and memory-aware. DeepSeek-R1-Distill-Qwen-14B’s docs benchmark the family with 32,768-token generation, so 8k is still well below the model’s advertised range. Intel’s notes also warn that users can set settings beyond their system’s ability, which reads like a hint that memory headroom, not the config file, is the real governor. If AI Playground is silently overriding the requested window, it should surface the effective cap and the reason instead of making the slider look authoritative.