Microsoft-Decision-1 (Azure Foundry) — A2 re-run
Data: JevBench v1.6.39, published 2026-10-22
Capability 72.2 · API offering, listed but not ranked on the API leaderboard
A2 supplement measurement with an incomplete equating pool (four of six); excluded from every ranking.
Last measured: 2026-10-12 · measurement release: v1.6.14
Licence not stated
intelligence
59.2
calibration
85.2
speed
88.7
cost
62.2
Composite (secondary)
71.5 · Not ranked on this board
Cost and measurement conditions
$0.018 per 1,000 decisions (not published)
LEAD DECISION r3: USD 0.042/M input and USD 0/M output, basis of the live Microsoft-Decision-1 (Azure Foundry) row on https://benchmarkheaven.com/jev-models/api. Reported usage across all 600 decisions, including seven HTTP422 refusals retained in denominator; frozen parent all-item cost basis. Invoice, warmup and refusal charges unknown.
p50 latency: 0.3 s
none (API)
Vendor-hosted Azure Foundry API; latency measured from Sandy (Germany). Seven of 600 HTTP 422 responses scored wrong. Four self-hosted systems re-run on H200, versus H100 parents; wide equating interval. MeTask 31B and Torchcast not completed after four failed pod attempts; see measurement notes.