Intern-Decision Multimodal Models for Structured JSON Outputs

Intern-Decision introduces 0.8B, 2B, and 4B multimodal models optimized for structured decision-making via autoregressive masked language modeling. The 4B model achieves 90.02% accuracy across seven benchmarks with a mean latency of 44.16 ms for 289-token requests on a single RTX 4090.