Decision 1.0: open-weight decision foundation models output typed probability distributions
kalyan_kpl · x · 2026-11-18
Decision 1.0 is an open-weight family of decision foundation models that turns textual states and runtime-defined questions into typed probability distributions, targeting reusable judgments inside software control flows.
- Choice, Noul and Score interfaces let apps specify outcomes at inference time and compose predictions into agents and workflows
- Each question yields a candidate distribution without autoregressive text generation
- Six models span bidirectional encoder and causal decoder architectures, including a workflow specialist
- Lux-9B scores 84.17% on explicit decisions and 91.25% on inference (weighted overall 77.22%); gaps remain in composition, transfer and probability calibration
Weights and inference code are released; paper: arXiv:2611.18629.
More from coding & agent
- Open-source MCP server connects Claude to Lychee Redmine project management — modelcontextprotocol · 2026-11-18
- Pharma Intel MCP chains ClinicalTrials.gov tools for AI pharma research — modelcontextprotocol · 2026-11-18
- Indie dev: skip GPT-6.2 hype, what I want is a better Antigravity harness — yihui_indie · 2026-11-18
- Microsoft launches Azure AI Gateway Labs: 27 hands-on labs for governing models and agents — adnan_hashmi · 2026-11-18
- Dev laments Codex downfall: great community, product no longer there — rudrank · 2026-11-18
- MCP overload bloats Claude Code context: dev shares a lean tool setup — PushZealousideal7914 · 2026-11-18