Open-source Karpo-Decision-31B tops 300-model Decision Index with zero-token, one-forward-pass answers
rohanpaul_ai · x · 2026-11-18
Karpo-Decision-31B, a new open-weight decision model, ties for #1 of 300 on the Decision Index Full board (73.9).
- Architecture: swaps decoding for a readout head — one token code per option, one score per code, softmax over the question's options. Zero generated tokens, always a valid answer, page shared across questions via branch attention.
- Benchmarks: built on Gemma 4 31B, it beats Google's Gemini Decision 1 on Amazon ESCI (64.7% vs 58.7%) and New Yorker Caption (90.5% vs 79.7%); JevBench 87.2 vs 78.5.
- Speed/cost: 95 ms median latency vs 196 ms for Jev 1.14, roughly a fifth cheaper per 1,000 decisions.
Suggested first use cases: search, recommendations, tool choice, model routing, browser actions.
More from Models
- Raschka: benchmarking new LLMs is like using IQ tests for hiring — rasbt · 2026-11-18
- GPT-6.2 backlash is real user experience, not astroturf: dev says Astra has quirks but Sol is a solid daily driver — pvncher · 2026-11-18
- LLM protein-folding test on fresh PDB entry lands near-random structures — MoAlQuraishi · 2026-11-18
- Every's founder pushes back: we re-tested the model to the wire, OpenAI gave us no heads-up — kieranklaassen · 2026-11-18
- HF trending: Thinking Machines' Inkling-2-Small, MiniMax H3-Image and Qwen3.8-27B surge together — victormustar · 2026-11-18
- US watchdog: four top AI chatbots helped plan terror attacks despite refusals — Dense-Following-9964 · 2026-11-18