Vega-U (Qwen3.5-4B reader)
Data: JevBench v1.6.39, published 2026-10-22
Capability 44.3 · Not ranked in the current release
latency 11.0× Jev (v1.5 reference)
Last measured: 2026-10-20 · measurement release: v1.6.32
vega-U checkpoint apache-2.0 (card and metadata agree); vegaml package apache-2.0; frozen reader Qwen/Qwen3.5-4B apache-2.0
intelligence
8.8
calibration
79.8
speed
60.7
cost
64.9
Composite (secondary)
0.8 · Composite rank #155
Cost and measurement conditions
$0.015 per 1,000 decisions (estimated)
ESTIMATE (*est.) through the frozen pricing_v15 path (module sha256 6202ef78...): vega-U is a 14.3 M decision engine on a FROZEN Qwen/Qwen3.5-4B reader, which resolves to deepinfra:Qwen/Qwen3.5-4B in the dated 25 Sep DeepInfra catalog snapshot (sha256 84eb7926acd4e93b71d695be24898e2ca999b20011d1b66cc0348ce7c6c3546f), USD 0.03/M input and USD 0.15/M output. The model generates no text: 0 output tokens on every item, so cost = input tokens x USD 0.03/M. COMMON COST BASIS: published cost is recomputed over the same remaining 1,485 items as every other row; all 1,500 items count for Intelligence. ALL-ITEMS DIAGNOSTIC: USD 0.0367598 per 1,000 over all 1,500 items (0.569x cap).
Measured p50 latency: 3.308 s (on-CPU, no network, loopback; not directly comparable with API latency, which we time end to end from our measuring host, network included)
Ranking and cap use: 6.77 s (est. incl. network (2x + 150 ms))
Self-hosted model: measured p50 3.308 s on-CPU, no network (timed on our CPU host over loopback). Not directly comparable with API latency, which we time end to end from our measuring host, network included. The ranking and latency cap use the same estimate for every self-hosted row: 2 × measured + 150 ms = 6.77 s. Model authors can ask us to time a hosted endpoint instead (on request).
x2 + 0.15 s (assumption, not measured)
Our own CPU server (AMD EPYC, 32 cores), disposable container with no network (cut proven from inside before the sealed run), reviewed vegaml==0.12.2, torch 2.14.1+cpu, transformers 5.19.0; fp32 throughout (the author's own backbone_dtype() choice on CPU); context 73,728, and a longer state is cut on the right by the author's own build_prompt: 15 of the 1,500 items sit at that cap and are answered on a truncated state. Reference PyTorch paths only.