IFM open-sources K2-Type-0.9B: a decision model outputting calibrated probabilities in one forward pass
HongyiWang10 · x · 2026-10-03
IFM's team showcases K2-Type-0.9B, an open decision model built on a 0.9B-parameter LLM. It takes one state plus many typed questions and outputs calibrated probabilities in a single forward pass—no text generated—balancing efficiency, effectiveness, and latency. Weights are Apache-2.0 on Hugging Face.
Advisor Hongyi Wang argues "big decisions don't always need big models," positioning it as an open-source exploration of compact, purpose-built decision models.
More from Models
- New proprietary VSA long-context solution claims up to +9.5 reasoning points at 100K tokens — teortaxesTex · 2026-10-03
- OpenAI's low-key Dot launch clashes with its collaborator-not-tool framing, critic argues — RileyRalmuto · 2026-10-03
- Sol 6.1 on the $500 Pro plan feels 'basically unlimited' in usage, user reports — soumitrashukla9 · 2026-10-03
- Claude Opus 5.5 renders elegant Chinese calligraphy via code and self-review — dotey · 2026-10-03
- Early users report GPT-6.1 Sol slashes usage consumption on identical Codex workloads — soumitrashukla9 · 2026-10-03
- Cohere Embed 5 Is First Model Family Benchmarked With New RCP-nDCG@10 Metric — cohere · 2026-10-03