Karpo-Decision-31Bv1.7 · 10 s rule
Data: JevBench v1.8.9, published 2026-11-11 · method v1.7 (10 s rule), measured under the rule
Capability 89.5 · 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: 2026-11-10 · measurement release: v1.8.9
Apache-2.0 (LICENSE in model and code repos); base google/gemma-4-31B-it Apache-2.0
intelligence
83.1
calibration
96.0
speed
88.3
cost
57.9
Composite (secondary)
78.4 · Not ranked on this board
Cost and measurement conditions
$0.025 per 1,000 decisions (estimated)
GPU-time estimate from v1.8.0: one monthly-rented RTX PRO 6000 96 GB, USD 1.5235/hour; average load 70%. Architecture: dense gemma, 31B active parameters; throughput 10106.516129 tokens/s (architecture-class calibration, efficient-engine prefill). GPU-time cost USD 0.0252347746738464/1,000; old list-price estimate USD 0.0379665815983882/1,000 (google/gemma-4-31b-it at USD 0.09/M input and USD 0.34/M output). Applied min(old list-price estimate, GPU-time cost) = USD 0.0252347746738464/1,000. Estimated, not a bill.
Measured p50 latency: 0.082 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: 0.31 s (est. incl. network (2x + 150 ms))
Self-hosted model: measured p50 0.082 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 = 0.31 s. Model authors can ask us to time a hosted endpoint instead (on request).
x2 + 0.15 s (assumption, not measured)
Evaluator-owned Verda GPU VM (RTX PRO 6000), offline read-only container (--network none); author's karpo_decision.serve.server unchanged at code commit 25c38826 (github.com/machinepulse-ai/karpo-decision), bf16 backbone + fp32 readout, one forward pass, no generation.
Measured latency over all decisions: p50 0.082 s · p95 0.161 s
— 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.
Model and serving disclosure
Paid evaluation, published with the author's consent. Measured by us on an evaluator-owned rented GPU with the author's own serving code at the pinned commit, offline (no network) during the sealed run. As agreed for this evaluation, this row shows the standard scores only; the per-category breakdowns are not published. Failed or unanswered requests count as wrong answers.