Jeff-1.1-Large (jgeuter)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 88.2 · 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
Weights CC BY-NC-4.0 per the model card (non-commercial; listed for evaluation only); base Apache-2.0; serving code Apache-2.0 (jeff 1.4.0).
intelligence
84.5
calibration
92.0
speed
72.2
cost
51.3
Composite (secondary)
71.3 · Not ranked on this board
Cost and measurement conditions
$0.042 per 1,000 decisions (estimated)
ESTIMATE (*est.), self-hosted open weights, no public tariff: base google/gemma-4-31B-it, google/gemma-4-31b-it at USD 0.09/0.34 per million input/output tokens (frozen pricing module sha256 645fd82eac728d597048bec0995ecd3ac896566e269fb6c9a1fdb84c171a9ac4). Native usage priced as reported: 620,215 input tokens (one-pass decision prompts after the author's long-input retrieval) and 21,830 output tokens (generated reasoning). The server does not count the two further prefills of a reasoned decision; our offline estimate of 199,718 such input tokens is disclosed, not priced (priced as input, the cost would be USD 0.054144 per 1,000 decisions). COST FIGURES: USD 0.042161 per 1,000 decisions over all 1,500 items of the draw (every row on this draw is costed on all 1,500 items); 0.65x the official cost cap USD 0.064595. Estimated model tariff, not charged usage or a host quote.
Measured p50 latency: 0.106 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: 0.36 s (est. incl. network (2x + 150 ms))
Self-hosted model: measured p50 0.106 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 = 0.36 s. Model authors can ask us to time a hosted endpoint instead (on request).
x2 + 0.15 s (assumption, not measured)
Measured 27 Oct 2026 on one rented Verda NVIDIA RTX PRO 6000 Blackwell Server Edition 96 GB (FIN-03, Finland, EU) over loopback. One measured attempt in a fresh process with pinned weights and serving code, read-only model and source, no credentials and networking disabled. The standard self-hosted x2 + 0.15 s latency adjustment applies. GPU rental is separate from the estimated model tariff.
Measured latency over 1,500 decisions: p50 0.106 s · p95 8.269 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. Adjusted p95 16.69 s, 0 decisions over the limit.