Minimodels Decision 1.0 (candy-cane)v1.7 · 10 s rule
Data: JevBench v1.8.11, published 2026-11-15 · method v1.7 (10 s rule), measured under the rule
Capability 82.5 · Not ranked in the current release
Added after the v1.8.10 ranking review; it is ranked at the next review.
Last measured: 2026-11-15 · measurement release: v1.8.11
apache-2.0 (model card); base google/gemma-4-31B-it (Gemma terms)
intelligence
74.2
calibration
90.7
speed
72.5
cost
57.6
Composite (secondary)
71.9 · Not ranked on this board
Cost and measurement conditions
$0.026 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 0.0258859962205545/1,000; old list-price estimate USD 0.0409948891873741/1,000 (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 0.0258859962205545/1,000. Estimated, not a bill.
Measured p50 latency: 0.130 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: 0.41 s (est. incl. network (2x + 150 ms))
Self-hosted model: measured p50 0.130 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 = 0.41 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 scripts/serve_torch_jev.py unchanged at model commit 14ea4e1d29571d58d1de224198de6babc78d7cd0. MLX 4-bit affine checkpoint dequantised to bf16 by the author's loader, with LoRA adapter and static candy_cane calibration. Author's selective scratchpad pass on uncertain choice items (greedy, capped at 180 tokens and 8.6 seconds); author's own long-input compression.
Measured latency over all decisions: p50 0.130 s · p95 6.827 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 public evaluation. Measured on the 27 Oct draw on an evaluator-owned rented GPU with the author's serving code unchanged at the pinned model commit, offline (no network). The author's wall-clock-capped selective scratchpad pass can make repeat runs differ slightly. Failed or unanswered requests count as wrong answers.