metask-jev-rain-31B-trainedv1.7 · 10 s rule
Data: JevBench v1.8.7, published 2026-11-03 · method v1.7 (10 s rule), re-scored from the recorded per-decision times
Capability 78.9 · rank #26 of 217 on the open-weights board
Last measured: 2026-10-12 · measurement release: v1.6.10
Weights: Apache-2.0 per model card (base Gemma 4 under the Gemma 4 licence); service code (service/) MIT
intelligence
70.7
calibration
87.1
speed
82.3
cost
56.1
Composite (secondary)
72.0 · Composite rank #44
Cost and measurement conditions
$0.029 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: dense gemma, 31B active parameters; throughput 10106.516129 tokens/s (architecture-class calibration, 29 Oct 2026, efficient-engine prefill; capped at 100,000 tokens/s). The row's own scored cost basis: 718,318 input + 1,485 output tokens over 1,485 decisions = 484.715824916 tokens/decision. GPU-time cost USD 0.0289953024393587/1,000; old list-price cost USD 0.0438744242424242/1,000; applied min(old list-price cost, GPU-time cost) = USD 0.0289953024393587/1,000. 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: 0.233 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: 0.62 s (est. incl. network (2x + 150 ms))
Self-hosted model: measured p50 0.233 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 = 0.62 s. Model authors can ask us to time a hosted endpoint instead (on request).
x2 + 0.15 s (assumption, not measured)
evaluator-owned Lium GPU pod (H100 80GB), offline read-only container (--network none); author's own service (service/serve.sh -> launch.py -> server.py on vLLM 0.31.0, pinned digest image), unchanged defaults paid evaluation (public). Own LoRA adapter (rank 16) on Google's official W4A16 Gemma 4 31B QAT checkpoint, applied unmerged by vLLM; max_model_len 6656, longer inputs are shortened by the service's documented extraction rule (author-tested only on H200; measured on H100 80GB with unchanged settings).
Measured latency over 1,500 decisions: p50 0.233 s · p95 0.399 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 78.91 · Composite 69.74 · rank #28 on the all-systems board · #12 open weights
intelligence 70.72 · calibration 87.09 · speed 83.18 · cost 50.73
v1.6 speed basis (124 items): p50 0.209 s · p95 0.348 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.