Qwen3.5-0.8B Decision Model (Mourad Ghafiri)v1.6 · re-measure pending
Data: JevBench v1.8.7, published 2026-11-03 · re-measure pending under the 10 s rule (v1.6 values)
Capability 42.6 · Not ranked in the current release
Re-measure pending under the 10 s rule: this row's per-decision times were recorded on a loaded CPU host, so they cannot be scored under the time limit. Shown with its v1.6 values until it is measured again.
Last measured: 2026-10-02 · measurement release: v1.6.0
MIT code and training data; Apache-2.0 model weights
intelligence
7.3
calibration
78.0
speed
55.9
cost
86.1
Composite (secondary)
0.5 · Not ranked on this board
Cost and measurement conditions
$0.0029 per 1,000 decisions (estimated)
GPU-time estimate from v1.8.0: one monthly-rented RTX PRO 6000 96 GB, EUR 999/month / 730 hours, ECB USD 1.1133/EUR dated 28 Oct 2026, USD 1.5235/hour; average load 70%. Architecture: small qwen, 0.8B active parameters; throughput 100000 tokens/s (architecture-class calibration, 29 Oct 2026, efficient-engine prefill; capped at 100,000 tokens/s). The row's own scored cost basis: 782,199 input + 0 output tokens over 1,624 decisions = 481.649630542 tokens/decision. GPU-time cost USD 0.00291187782591485/1,000; old list-price cost USD 0.00481649630541872/1,000; applied min(old list-price cost, GPU-time cost) = USD 0.00291187782591485/1,000. The GPU-time estimate is lower. Estimated, not a bill. Intelligence, Calibration, measured latency and the common cost-item membership are unchanged. Prior prices remain in the archived v1.7 releases.
Measured p50 latency: 5.551 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: 11.25 s (est. incl. network (2x + 150 ms))
Self-hosted model: measured p50 5.551 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 = 11.25 s. Model authors can ask us to time a hosted endpoint instead (on request).
x2 + 0.15 s (assumption, not measured)
evaluator-owned CPU container, offline read-only, pinned cores
not yet re-scored under the 10 s rule.
JevBench v1.6.43 values (method v1.6, before the 10 s rule)
Re-measure pending under the 10 s rule: this row's per-decision times were recorded on a loaded CPU host, so they cannot be scored under the time limit. Shown with its v1.6 values until it is measured again.
Capability 42.63 · Composite 0.47 · rank #207 on the all-systems board · #164 open weights
intelligence 7.30 · calibration 77.96 · speed 55.91 · cost 79.52
v1.6 speed basis (124 items): p50 5.551 s · p95 11.327 s