ResearchBenchmarksNavigation Visual Question Answering on NaVQA Long (L) horizonFollow80Descriptive AccuracyRAVEN64.468.4572.576.55Jun 23, 2026Evaluation ResultsMethodMethodLinksDescriptive AccuracyPositional Error (m)Temporal Error (min)RAVENLLM=Gemini-3-Pro, Enco...LLM=Gemini-3-Pro, Encoder=QQMM-v2, Quantization=4-bit2026.068038.810.07VLM OnlyLLM=GPT-4oLLM=GPT-4o2026.066540.332.68