Internal Productllm
Continuum1-9B
Long-context foundation model with hybrid linear attention — open weights on Hugging Face.
Params: ~8.6BContext: 2M tokensMMLU: ~75% (protocol)
Research
Research you can run.
Programs with delivery layers you can inspect, public model releases, and evaluation practice that feeds client engineering — not invented peer-review claims.
Research-grade evaluationProduction-minded engineeringResponsible handover
Research programs
Each program covers evaluation, model development, inference, and integration — with public releases as a concrete starting point.
Evidence note: Releases below are Innomium internal products and technical demonstrations unless explicitly labeled otherwise. Named client case studies appear only with written publication approval.
Public releases
Methods & notes
All insights- visionBuilding the Innomium Vision LayerHow we turn existing camera infrastructure into real-time intelligence — without new hardware, cloud lock-in, or heavyweight models.
- visionSentinel Hits 92% on Crowd ScenesOur edge person detector now holds 92% accuracy on dense airport and stadium footage — in a 19MB ONNX package with a live browser demo.
- visionEmber: Outdoor Fire Detection for Mission SitesFire and early-smoke detection tuned for outdoor fuel, logistics, and perimeter monitoring — now shipping with live demos and an active builder challenge.
- visionVantage Reaches 93% on Highway ScenesVehicle detection optimized for toll lanes, merge points, and low-light traffic — compact enough for edge gates and parking operations.
Discuss a research-led engagement.
Scoped R&D, model adaptation, or production integration — start with the operating constraint.