BB-Qwen3.5-4B-LoRA (Babak Barazandeh, LoRA on Qwen3.5-4B-Base, hosted /v1/systemone)v1.7 · 10 s rule

Data: JevBench v1.8.18, published 2026-11-20 · method v1.7 (10 s rule), re-scored from the recorded per-decision times

Capability 60.4 · API offering, listed but not ranked on the API leaderboard

registry ranked=false or no composite

Last measured: 2026-10-07 · measurement release: v1.6.1

Licence not stated

Base model: Qwen/Qwen3.5-4B-Basesource

intelligence

38.7

calibration

82.1

speed

80.9

cost

—

Composite (secondary)

— · Not ranked on this board

Cost and measurement conditions

not published (not published)

UNPRICED - NOT A MEASUREMENT FAILURE. The base is disclosed (Qwen/Qwen3.5-4B-Base; reference() resolves it to deepinfra:Qwen/Qwen3.5-4B at 0.03/0.15), but the endpoint reports usage.input_tokens ONLY - no output_tokens on any answer - and the system is API-only, so 'nothing is generated' cannot be verified in its code; the published 02 Oct rule keeps an unverified null unpriced. The frozen scorer therefore gives no Cost axis and no composite; the row is reported on Capability. PRICING REFERENCE NOTE: if the author confirms (or independently verified evidence establishes) a one-pass readout with 0 output tokens, the row prices by one multiplication from the raw at 0.03/M input (receipts/COST-ALTERNATIVES-R59.json).

p50 latency: 0.7 s

none (API)

Hosted full-set API measured 2026-10-07; submitted LoRA checkpoint-030500 on Qwen3.5-4B-Base, weights not released. Serial requests over author ngrok tunnel, no retry; latency includes transit. Complete 1500/1500 answers. Input usage is reported but output usage is absent, so Cost and composite remain unpriced; Capability is reported. Exposure ledger recorded before dispatch.

Measured latency over 1,500 decisions: p50 0.717 s · p95 1.131 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.

JevBench v1.6.43 values (method v1.6, before the 10 s rule)

Capability 60.41 · Composite —

intelligence 38.67 · calibration 82.14 · speed 81.58 · cost —

v1.6 speed basis (124 items): p50 0.695 s · p95 1.000 s

These figures are kept for comparison. The current figures above are re-scored under the 10 s rule from the recorded per-decision times and replace them.

Published source