Decision models SOTA: vLLM-SR d3 (27B) & linne-oak 1.0 (31B) &…

CategoryDecision models· Decision models

Also in Decision models:VisionC

Decision models · Oct 14, 2026

Who is SOTA in Decision models?

vLLM-SR d3 (27B) & linne-oak 1.0 (31B) & Perplexity Decider v1.1 27B (Qwen3.8-27B, noncausal decision readout)

Typed answers with probabilities. The crown follows Decision Index Full and JevBench, both independent at weight 1. Vendor-reported numbers are display-only at weight 0. As of Oct 14, 2026, vLLM-SR d3 (27B) & linne-oak 1.0 (31B) & Perplexity Decider v1.1 27B (Qwen3.8-27B, noncausal decision readout) are tied for state of…

Overall Open weights Calibration Latency Price Hosted ≤ 32B Best per $

Standings · overall

The board

RSS History →

#	Model	State	Confidence	Why
1	VLvLLM-SR d3 (27B) vLLM Semantic Router	Co-SOTA · Decision models	low	vLLM-SR d3 (27B) is co-SOTA in Decision models with low confidence because primary sources disagree beyond the margin of error (with…
1	MIlinne-oak 1.0 (31B) MISA-AICORE	Co-SOTA · Decision models	low	linne-oak 1.0 (31B) is co-SOTA in Decision models with low confidence because primary sources disagree beyond the margin of error (with…
1	PEPerplexity Decider v1.1 27B (Qwen3.8-27B, noncausal decision readout) Perplexity	Co-SOTA · Decision models	low	Perplexity Decider v1.1 27B (Qwen3.8-27B, noncausal decision readout) is co-SOTA in Decision models with low confidence because primary…
2	UNDecider 31B (NVFP4) Unknown	Near-SOTA · Decision models	medium	Close in Decision models: rank #2, 1.2 percentile points behind the leader.
3	STSieve 31B Sthānika AI	Near-SOTA · Decision models	medium	Close in Decision models: rank #3, 1.8 percentile points behind the leader.
4	AMdecisio v0.8.0 on gemma-4-31B-it (frozen, FP8 on load, one prefill per state, self-hosted) Amin Roudaki (@aminry)	Near-SOTA · Decision models	medium	Close in Decision models: rank #4, 2.3 percentile points behind the leader.
5	UNBobcat 1.3 (31B) Unknown	Near-SOTA · Decision models	medium	Close in Decision models: rank #5, 2.4 percentile points behind the leader.
6	FIDarwin-27B-ZTC FINAL_Bench	Near-SOTA · Decision models	medium	Close in Decision models: rank #6, 3.6 percentile points behind the leader.
7	UNMORJ 1.0.1 (Gemma 4 31B) Unknown	#7 in Decision models	medium	Ranked #7 in Decision models.
8	TOtorchcast-decision-27b (Torchcast AI, Qwen3.8-27B fine-tune) Torchcast AI	#8 in Decision models	medium	Ranked #8 in Decision models.
9	UNEikos-27B (caiovicentino1, Qwen3.8-27B) Unknown	#9 in Decision models	medium	Ranked #9 in Decision models.
10	UNAnyJev L0 · Gemma-4-31B Unknown	#10 in Decision models	medium	Ranked #10 in Decision models.
11	ELRené-1 31B FP8 (salfatigroup, Gemma 4 31B full fine-tune, one-pass option readout) Elon Salfati / salfatigroup	#11 in Decision models	medium	Ranked #11 in Decision models.
12	FAFastino GLiDE no-thinking (28B) Fastino	#12 in Decision models	medium	Ranked #12 in Decision models.
13	UNQuyet-1.0-Large (Chinh Nguyen, Gemma-4-31B decoder, option-letter logits) Unknown	#13 in Decision models	medium	Ranked #13 in Decision models.

Embed

Badges for the leaders

vLLM-SR d3 (27B)

[![CO-SOTA · Decision models · Oct 2026](https://isitsota.com/badge/vllm-sr-d3-v310/decision.svg)](https://isitsota.com/m/vllm-sr-d3-v310)

linne-oak 1.0 (31B)

[![CO-SOTA · Decision models · Oct 2026](https://isitsota.com/badge/linne-oak-1-0/decision.svg)](https://isitsota.com/m/linne-oak-1-0)

Perplexity Decider v1.1 27B (Qwen3.8-27B, noncausal decision readout)

[![CO-SOTA · Decision models · Oct 2026](https://isitsota.com/badge/pplx-decider-v1-1-27b/decision.svg)](https://isitsota.com/m/pplx-decider-v1-1-27b)

Calibration

Reliability · Nace.AI Drex v2.0 (35B-A3B)

Confidence on the horizontal axis, accuracy on the vertical axis. Larger marks carry more weight.

Display only

Self-reported claims

No vendor-reported row is in this snapshot. Vendor-reported numbers are display-only at weight 0.

Best page and recent changes

Best AI model for decision models

What is a decision model

Evidence recipe

What can move this crown

Decision Index Full — primary, weight 1
JevBench — primary, weight 1
Decision Index latency — display, weight 0 (display only, cannot crown)
JevBench price — display, weight 0 (display only, cannot crown)
Vendor decision claims — display, weight 0 (display only, cannot crown)

Who is SOTA in Decision models?

As of Oct 14, 2026, vLLM-SR d3 (27B) & linne-oak 1.0 (31B) & Perplexity Decider v1.1 27B (Qwen3.8-27B, noncausal decision readout) are tied for state of the art with low confidence in Decision models (since Oct 14, 2026; sources Decision Index Full and JevBench; methodology v1.0).

Since when is the Decision models crown current?

The current crown in Decision models has held since Oct 14, 2026.

What is a decision model?

A decision model takes state, as text or as images, plus a typed question. It returns a typed answer with probabilities rather than free text. Output types: noul (yes/no/unknown), choice, score, multi-label.

How is calibration ranked?

The calibration slice ranks the lowest ECE among models in the top quartile of accuracy, so a constant probability cannot take the slice. Brier breaks ECE ties.

Do vendor-reported decision numbers count?

Vendor-reported numbers are display-only at weight 0.