Shanghai AI Lab open-sources Intern-Decision models from 0.8B to 4B

Shanghai AI Lab has open-sourced Intern-Decision, a family of compact decision models at 0.8B, 2B and 4B parameters that return structured judgments with probabilities instead of text. The 4B model scores 90.02% average accuracy across seven benchmarks and runs at about 44 ms on a single RTX 4090.

• Three sizes: 0.8B, 2B and 4B parameters, weights on Hugging Face
• Question types: choice, score on a custom scale, and yes/no with probabilities
• 4B model: 90.02% accuracy vs 88.74% for Jev across seven suites
• About 44 ms latency on a single consumer RTX 4090