Google launches Gemini Decision 1: a decision model that outputs probabilities, not text
ariG23498 · x · 2026-11-12
- Gemini Decisions API / Gemini Decision 1 announced: a new "decision model" paradigm from Google with multimodal input (audio, video, PDF), 256k context, and 150-250ms end-to-end latency.
- Core idea: LLMs excel at generating text, but many tasks (e.g. customer-support triage) don't need it. Instead of generating JSON, a decision model directly evaluates the available options and returns a probability distribution over them.
- Hugging Face primer: the HF team published a tutorial showing how to call decision models via a transformers pipeline (e.g. autotrust/GEV-26B-Decide), using a support-ticket routing example to demonstrate defining choice-type questions and criteria.
The paradigm targets high-throughput, low-latency routing/classification workloads where text generation is overhead.
Related event: Google launches Gemini Decisions API with Decision 1 model(32 posts)→
More from Models
- Ben's Bites: Anthropic builds office suite inside Claude, Gemini 4 Carbon rolls out — Ben's Bites · 2026-11-12
- Unsloth's new quantization makes local Qwen Flash Next feel almost like Qwen4 — danielhanchen · 2026-11-12
- Benchmark criticized for cherry-picking known failures; same model scores 75% vs 56% across harnesses — tallmetommy · 2026-11-12
- RF-DETR Large: best OSS-licensed object detector at just 34M params — TheZachMueller · 2026-11-12
- Telling Gemini you're Karpathy makes it think 30% longer and score 7.8 vs 5.8 in blind test — dejanseo · 2026-11-12
- Continu Intel launches ci-proto: tops fev-bench with 64x context window, open weights — s_mohinii · 2026-11-12