Instinct Dual 4Bv1.7 · 10 s rule
Data: JevBench v1.8.11, published 2026-11-15 · method v1.7 (10 s rule), re-scored from the recorded per-decision times
Capability 57.2 · API offering, ranked on the API leaderboard · Capability rank #33 of 39 within the Jev-class caps
Measured 2026-10-06 on A4 ∪ P, 600 items, equated (+2.57 I / +3.35 C); round score-a4-2.
Not independently recorded
intelligence
28.3
calibration
86.1
speed
88.6
cost
75.0
Composite (secondary)
18.0 · Composite rank #44 of 54 on the API leaderboard
Cost and measurement conditions
$0.0068 per 1,000 decisions (public tariff)
measured: token usage of the 6 Oct 2026 A4 u P run x tariff USD 0.01/M input, 0/M output (spec instinct-dual-4b.json), per 1,000 decisions on the v1.6.1 common cost basis (591 of 600 items; the 9 A4 u P items outside the Jev reference's accepted input range are excluded, cost-basis-a4.json)
p50 latency: 0.2 s
none (API)
Measured latency over 600 decisions: p50 0.232 s · p95 0.601 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.
JevBench v1.6.43 values (method v1.6, before the 10 s rule)
Capability 57.21 · Composite 17.97 · rank #142 on the all-systems board · #36 API
intelligence 28.34 · calibration 86.08 · speed 88.60 · cost 75.03
v1.6 speed basis (124 items): p50 0.231 s · p95 0.598 s
These figures are kept for comparison. The current figures above are re-scored under the 10 s rule from the recorded per-decision times and replace them.
Breakdowns (not part of the score)
Chance-corrected competence (0 = chance, 100 = perfect) on A4 ∪ P (600 items: 300 open + 300 sealed; fewer sealed items than the full set, so wider uncertainty). Raw values, not equated. Category cells use A4+P+L1+L2+L3; per-type/tier cells use the original basis above. Radars against Jev.
| Request type | Open | Sealed |
|---|---|---|
| choice | 51.7 | 49.0 |
| noul | 0.0 | 0.0 |
| score | 27.1 | 26.7 |
Capability by subject topic
| Rules, policy & law | 4.5 | n=1133 |
|---|---|---|
| Coding & software | 31.6 | n=202 |
| Math & numbers | 0.0 | n=269 |
| Finance & commerce | 5.9 | n=178 |
| Support & operations | 27.4 | n=109 |
| Everyday language | 52.2 | n=130 |
| Safety & security | 27.4 | n=84 |
Use cases (TypeSafe categories)
| Model routing | 43.9 | n=267 |
|---|---|---|
| Legal & compliance | 17.3 | n=269 |
| Customer support | 12.3 | n=156 |
| E-commerce | 4.0 | n=88 |
| Insurance claims | 9.9 | n=92 |
| Risk assessment | 7.2 | n=88 |
| Financial crime | 0.7 | n=76 |
| Feature extraction | 0.0 | n=90 |
| Lead generation | 14.9 | n=86 |
| Recruiting | 0.0 | n=88 |
| Knowledge graphs | 0.0 | n=82 |
| LLM guardrails | 6.7 | n=82 |
| Moderation | 5.6 | n=69 |
| Code linting | 0.0 | n=81 |
| Search & retrieval | 35.4 | n=84 |
| Science | 0.0 | n=82 |
| Advertising | 0.0 | n=85 |
| Gaming | 0.0 | n=79 |
| Demand forecasting | 0.0 | n=72 |
Cells need at least 30 answered items; † fewer than 30 (indicative only; drawn as hollow, lighter points on the compare radars).
Previous measurement (v1.5.x, not comparable)
Capability 57.2 · composite 18.0 · cost $0.0068 per 1,000 decisions (public list price)
Last measured 2026-10-06, measurement release v1.6.1. Older method and scale; the current figures above replace it.