Bev / Bonsai 27B
system-one-open · by Reza Sayar · Code and weights marked open
Base model: Qwen/Qwen3.8-27Bsourcev1.5 roster addendum A4
Not independently recorded
JevBench v1.5.4 score
15.767
Option A: rank #61 of 106 ranked systems.
The three option scores and ranks are published independently; the headline is Option A.
| Option | Score | Rank |
|---|---|---|
| A · headline | 15.767 | #61 of 106 |
| B | 15.985 | #61 of 106 |
| C | 12.339 | #62 of 106 |
Published axes
- intelligence
- 53.1
- calibration
- 77.7
- speed
- 72.7
- cost
- 28.2
Bands use the published 0–100 axis values; the marked reference is Jev 1.13.0 when that axis is available.
Run and cost evidence
- Run status
- complete · 1,624 decisions · 0 missing
- Cost
- $0.247 per 1,000 decisions · estimate · ESTIMATE: documented hosted-model estimate. Frozen market reference qwen/qwen3.8-27b.
- Median latency
- 2.191 seconds, adjusted · x2 + 0.15 s (assumption, not measured)
- Endpoint condition
- evaluator-owned H100 GPU
- Published source
- https://huggingface.co/Reza2kn/Bev
Model and serving disclosure
Reza2kn/Bev @ f01edf952e3f2da8147aa08ec6d6229c863d5d19. Unchanged Prism ML Ternary-Bonsai-2-27B PQ2_0 quantization of Qwen3.8-27B, with selected-token readout; not a fine-tune. GGUF SHA-256 3907dc1658db1f78a9826bf8d5bcb8dc65db0d466388937af57f2294fae62ec1. H100; CUDA backend compiled from the author-pinned source 9a9394a895b96003ca842a6041cb28ac49a108f7 with the author patch, rather than their prebuilt binary. The wire shim drops noul criteria because the submitted schema rejects them; that loses option descriptions for noul.
Values come from the public v1.5.4 aggregate. Scores and ranks may change in a later release.
Read the full leaderboard, the v1.5.4 release page, and the published method.