SubQ Snap 1.2v1.7 · 10 s rule
Data: JevBench v1.8.18, published 2026-11-20 · method v1.7 (10 s rule), measured under the rule
Capability 83.8 · API offering, ranked on the API leaderboard
Last measured: 2026-11-12 · measurement release: v1.8.17
Proprietary hosted API (no weights release)
intelligence
75.5
calibration
92.2
speed
87.0
cost
64.8
Composite (secondary)
78.4 · Not ranked on this board
Cost and measurement conditions
$0.015 per 1,000 decisions (estimated)
SubQ's published launch tariff, USD 0.04 per million input tokens and free output (https://subq.ai/pricing, read 12 Nov 2026). Counted under the frozen I-1 launch-list-price rule, excluding promotional credits. The common 1,489-item basis uses 555,139 measured input tokens and yields USD 0.0149130691739422 per 1,000 decisions. The full 1,500-item run used 952,081 input tokens and yields USD 0.0253888266666667 per 1,000 decisions. Both figures use zero output tokens.
p50 latency: 0.4 s
none (API)
SubQ /v1/systemone API (api.subq.ai), one item per request, native probabilities; latency measured from our server in Germany.
Measured latency over 1,500 decisions: p50 0.436 s · p95 0.453 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.
Breakdowns (not part of the score)
Chance-corrected competence (0 = chance, 100 = perfect) by category on open+sealed. Raw values, not equated; the same cells feed the compare view.
Capability by subject topic
subject topics: not labelled for this draw yet (top-up pending)
Use cases (TypeSafe categories)
| Model routing | 83.0 | n=148 |
|---|---|---|
| Legal & compliance | 78.6 | n=56 |
| Customer support | 73.4 | n=66 |
| E-commerce | 75.3 | n=62 |
| Insurance claims | 71.3 | n=62 |
| Risk assessment | 71.1 | n=48 |
| Financial crime | 64.1 | n=63 |
| Feature extraction | 81.8 | n=72 |
| Lead generation | 88.5 | n=75 |
| Recruiting | 96.3 | n=65 |
| Knowledge graphs | 94.8 | n=67 |
| LLM guardrails | 95.2 | n=71 |
| Moderation | 94.8 | n=59 |
| Code linting | 91.9 | n=69 |
| Search & retrieval | 99.4 | n=67 |
| Science | 54.7 | n=42 |
| Advertising | 60.8 | n=48 |
| Gaming | 73.8 | n=58 |
| Demand forecasting | 16.0 | n=66 |
Cells need at least 30 answered items; † fewer than 30 (indicative only; drawn as hollow, lighter points on the compare radars).
Model and serving disclosure
Measured through the SubQ hosted API on the full 27 Oct draw. Hosting implementation and deployed weights were not independently inspected. Failed or unanswered requests count as wrong answers. The hosted API returned native probabilities for all three question types and reported zero output tokens.