Jev-Style-2Bv1.7 · 10 s rule
Data: JevBench v1.7.23, published 2026-10-29 · method v1.7 (10 s rule), measured under the rule
Capability 61.2 · Not ranked in the current release
Measured on the 27 Oct 2026 draw, which is not yet equated to the reference scale; stronger anchor runs for that draw are being measured. Shown unranked until then.
Last measured: 2026-10-28 · measurement release: v1.7.17
Apache-2.0
intelligence
33.9
calibration
88.5
speed
94.4
cost
75.4
Composite (secondary)
28.5 · Not ranked on this board
Cost and measurement conditions
$0.0066 per 1,000 decisions (estimated)
ESTIMATE (*est.), base-model market reference Qwen/Qwen3.5-2B through the frozen pricing module's reference(): deepinfra:Qwen/Qwen3.5-2B, USD 0.02 per million input tokens and USD 0.10 per million output tokens (frozen pricing module sha256 645fd82eac728d597048bec0995ecd3ac896566e269fb6c9a1fdb84c171a9ac4). Native usage: 491,898 input tokens and 0 output tokens on the 1,489 common items; output 0 as reported, with no nominal token added. COST FIGURES: USD 0.006607 per 1,000 decisions, 0.10x the official cost cap USD 0.064595. Cost on the 1,489 items every row on this draw answered (same-items rule); all-items cost USD 0.006589 per 1,000 decisions. Estimated model tariff, not charged usage or a host quote; GPU rental is separate.
Measured p50 latency: 0.013 s (on-GPU, 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: 0.18 s (est. incl. network (2x + 150 ms))
Self-hosted model: measured p50 0.013 s on-GPU, no network (timed on our GPU 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 = 0.18 s. Model authors can ask us to time a hosted endpoint instead (on request).
x2 + 0.15 s (assumption, not measured)
Self-hosted on one evaluator-rented Verda RTX 6000 Ada GPU in Finland (EU), through the author's own documented HTTP server: CUDA, bf16, CUDA graphs and FLA kernel. Measured 28 Oct 2026, 16:45-16:56 UTC. Contributor code ran only on a disposable rented pod, with network cut before the scored run. GPU rental is separate from the estimated model tariff.
Measured latency over 1,500 decisions: p50 0.013 s · p95 0.027 s
0 decisions over 10 s.
A decision that takes longer than 10 seconds (measured time) counts as a wrong answer and as zero calibration for that item: each type's calibration is multiplied by the share of its decisions answered within 10 s. The limit uses the measured time (raw). The displayed adjusted time for self-hosted rows (2 × + 0.15 s) is an assumption and is not used for the limit.