Decisio 12B tuned commit-think (variant of the Decisio family)v1.7 · 10 s rule
Data: JevBench v1.7.14, published 2026-10-28 · method v1.7 (10 s rule), measured under the rule
Capability 78.7 · Not ranked in the current release
Measured on the 27 Oct 2026 draw, which is not yet equated to the reference scale. This draw differs strongly from earlier draws (72% multilingual items). Shown unranked until the draw is equated.
Last measured: 2026-10-27 · measurement release: v1.7.14
Apache-2.0 per the model card, LICENSE and NOTICE on the weights and the author server; base google/gemma-4-12B-it public, ungated, under its own terms
intelligence
68.6
calibration
88.7
speed
90.6
cost
56.1
Composite (secondary)
73.1 · Not ranked on this board
Cost and measurement conditions
$0.029 per 1,000 decisions (estimated)
ESTIMATE (*est.), self-hosted open weights, no public tariff: the declared base google/gemma-4-12B-it has no listing in the frozen pricing module (sha256 645fd82eac728d597048bec0995ecd3ac896566e269fb6c9a1fdb84c171a9ac4), so the settled Gemma-4-12B estimate applies: the nearest listed 12B Gemma sibling, google/gemma-3-12b-it at USD 0.05/0.15 per million input/output tokens. Billed on engine-processed input: the primary prompt the engine reads per decision, chat template and system turn included, 844,031 input tokens on all 1,500 items by an exact offline per-item recount (0 mismatches; decisio's own usage reports 727,045 tokens without the template and system turn, 78 tokens less per item, 77 on 14 items at a tokenizer merge seam); the 5 cut items count at their cut size. Output: 9,121 tokens as reported natively (1,500 option-label readouts plus 7,621 reasoning tokens from the 36 decisions that thought), priced at the full output rate. Not priced: the second read of the think pass, which has no native count (the run thought on 36 items; an offline upper bound for its prefill, if all 746 think-eligible items had thought, is 504,719 tokens), and any state-prefix registration prefill. COST FIGURES: USD 0.029046 per 1,000 decisions over all 1,500 items of the draw (every row on this draw is costed on all 1,500 items); 0.45x the official cost cap USD 0.064595. Estimated model tariff, not charged usage or a host quote.
Measured p50 latency: 0.052 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: 0.25 s (est. incl. network (2x + 150 ms))
Self-hosted model: measured p50 0.052 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 = 0.25 s. Model authors can ask us to time a hosted endpoint instead (on request).
x2 + 0.15 s (assumption, not measured)
Evaluator-owned rented Verda GPU VM, one NVIDIA RTX PRO 6000 Blackwell Server Edition 96 GB (97,887 MiB, driver 580.178.04), FIN-03 (Finland, EU), PAYG non-spot, instance type 1RTXPRO6000.30V, image 24.04.cuda13.0, USD 2.608 per hour recorded by the runner, 150 GB disk. One VM for this row, one attempt. Measured 27 Oct 2026, VM window about 17:35-17:55Z, about USD 0.8 GPU rental (elapsed time at the recorded hourly rate). Author server at 127.0.0.1:8000, network cut before sealed input (offline proof and cut proof receipts), read-only model and input, private output, HF offline flags, no credentials. Before sealed input: the 20 model files re-hashed against SHA256SUMS on the node and in the run container; the dependency Triton gate, a pre-warm that asserted the full served configuration on the scored process, and the smoke passed; the TTL fit gate admitted the run. Input sha256 9b2f32edaccb29afd726af8cd54cf1a22d85e8b1824566924045d83c6f828426 (v1.6-regular-20261027) was re-checked inside the container (match). Complete 1,500-row run, rc 0. Raw output sha256 d7396dd680e13af9ba26c4526007fee3647d03eee7673bf88075ac510f39cc3c (the pull-hash receipt). Node-shred and local-shred receipts verified; exact-id teardown receipt confirms the VM closed. GPU rental is separate from the benchmark's estimated model tariff.
Measured latency over 1,500 decisions: p50 0.052 s · p95 0.096 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.