Shanghai AI Lab Open-Sources Intern-Decision Model Family
Shanghai AI Lab's InternLM team released Intern-Decision, a family of multimodal decision models at 0.8B, 2B and 4B, open-sourced under Apache 2.0 with weights on Hugging Face, GitHub and ModelScope. Built on Qwen3.5, the models target multimodal decision-making with calibrated probabilities.
Confirmed
- Designed for structured decisions: supports choice, scoring, and yes/no question formats, handling up to 16 decision questions per request
- The three models average 79.38 / 84.68 / 90.02 across seven decision benchmarks
- According to the releasers (@Spico197, @stingning), the 4B version beats Jev 1.13.0 in effectiveness, speed, and probability calibration
Why it matters
- Small 0.8B–4B models focused on structured decision-making, combined with the permissive Apache 2.0 license, lower the barrier to adopting and building on multimodal decision capabilities
- Compact models achieving near- or above-par benchmark performance offer a lightweight option for decision-centric applications
2026-09-28 ~ 2026-09-30 · 5 related posts
Primary sources
- [source] InternLM open-sources Intern-Decision: 4B multimodal model beats Jev 1.13.0 — Spico197 · 2026-09-27
- Shanghai AI Lab ships Intern-Decision multimodal family, 4B model beats Jev 1.13.0 — stingning · 2026-09-28
- [source] Shanghai AI Lab releases Intern Decision: 0.8B-4B Apache 2.0 models for structured decisions — AdinaYakup · 2026-09-29
- Intern-Decision open models (0.8B–4B) beat Jev on multimodal decision benchmarks — max_paperclips · 2026-09-30
1 near-duplicate retellings: AdinaYakup