Ling 3.0 Tiny: community MLX local-agent qualification receipt
Sharing a scoped local-agent serving reference for Ling 3.0 Tiny. This adds evidence for a Rapid-MLX route alongside the local deployment options already documented by the project.
The September qualification used ling-3.0-tiny-4bit, the glm47 tool parser and exact-build qualification ling-3.0-tiny-4bit-v1. The committed receipt records M3 Ultra 256 GB, macOS 26.5.2, Rapid 0.14.2 source fc3807015, MLX 0.32.2, mlx-lm 0.31.3 and Python 3.12.14.
The five-task, three-seed harness passed 15/15, covering context recall, tool restraint, weather, search/browse and prompt-injection handling. This is our bounded product harness with controlled tools, not a general agent benchmark or a guarantee of safe tool behavior. The report links the JSON receipt and explains exact model/parser matching and held-out builds.
The Personal Intelligence feature and its qualification registration were retired on 2026-10-18; this is historical evidence, not a claim that those product gates still exist. It may inform a new Apple Silicon community recipe; any updated checkpoint or serving configuration should be qualified separately before inheriting this result.
Evidence and reproduction details: checked-in qualification report. These are historical results, not a new inference run or an upstream endorsement.
Disclosure: I maintain Rapid-MLX; this contribution was prepared with AI assistance under my direction. Anonymous usage-metadata telemetry is enabled by default; DO_NOT_TRACK=1 disables it. Model and conversion licenses still apply.