Is vLLM-SR d3 (27B) SOTA in Decision models?
Co-SOTA · Decision modelsvLLM-SR d3 (27B) is co-SOTA in Decision models with low confidence because primary sources disagree beyond the margin of error (with linne-oak 1.0 (31B), Perplexity Decider v1.1 27B (Qwen3.8-27B, noncausal decision readout)).
Standings
Decision models
| # | Model | State |
|---|---|---|
| 1 | vLLM-SR d3 (27B) | Co-SOTA |
| 1 | linne-oak 1.0 (31B) | Co-SOTA |
| 1 | Perplexity Decider v1.1 27B (Qwen3.8-27B, noncausal decision readout) | Co-SOTA |
| 2 | Decider 31B (NVFP4) | Near-SOTA |
| 3 | Sieve 31B | Near-SOTA |
| 4 | decisio v0.8.0 on gemma-4-31B-it (frozen, FP8 on load, one prefill per state, self-hosted) | Near-SOTA |
| 5 | Bobcat 1.3 (31B) | Near-SOTA |
| 6 | Darwin-27B-ZTC | Near-SOTA |
| 7 | MORJ 1.0.1 (Gemma 4 31B) | #7 in this category |
| 8 | torchcast-decision-27b (Torchcast AI, Qwen3.8-27B fine-tune) | #8 in this category |
| 9 | Eikos-27B (caiovicentino1, Qwen3.8-27B) | #9 in this category |
| 10 | AnyJev L0 · Gemma-4-31B | #10 in this category |