Shanghai AI Lab Open-Sources Intern-Decision, Small Models That Output Decisions, Not Text

Shanghai AI Lab released the Intern-Decision models on September 28, available in three sizes: 0.8B, 2B, and 4B parameters. These models are designed to provide structured judgments directly rather than generating long text passages, offering probabilities for each possible answer. Intern-Decision supports three question types: choice questions, score questions, and yes/no questions. A single request can bundle several questions about the same state, all answered in one forward pass. The models fine-tune the language backbone of Qwen3.5 while keeping the vision components frozen.