Standard One 3B SH (Standard Thinking)
Data: JevBench v1.6.40, published 2026-10-22
Capability 55.6 · rank #79 of 139 on the open-weights board
Last measured: 2026-10-12 · measurement release: v1.6.10
Apache-2.0 (merged weights, schema head and reference server; base Ministral 3 3B Apache-2.0)
intelligence
32.8
calibration
78.4
speed
86.6
cost
48.6
Composite (secondary)
21.6 · Composite rank #85
Cost and measurement conditions
$0.052 per 1,000 decisions (estimated)
v1.5-M2 base-model reference (read 2026-10-12T00:14Z, source/public-docs/OPENROUTER-MINISTRAL-3-PRICES.json): the model card names base_model StandardThinking/StandardOne-3B, built on mistralai/Ministral-3-3B-Instruct-2512. Neither Standard One checkpoint is listed on OpenRouter; the exact underlying base is: OpenRouter mistralai/ministral-3b-2512 list price USD 0.10/M input and USD 0.10/M output. Self-hosted weights, no bookable tariff of the measured deployment; the author's stated hosted list price (USD 0.038/M input, standardthinking.ai/pricing) is below this floor, so the M2 §2 floor binds (price floor: base-model reference price applied). The schema head generates no text, so output tokens are 0 and priced at the reference output price. COMMON COST BASIS (v1.6.30): cost per 1,000 decisions is computed on the common 1,485 items of this draw, leaving out the 15 very long items (at least 76,358 input tokens as counted by the rows that answered them). They are the 15 items the Jev reference refused with max_tokens_exceeded, which the other rows that did not answer them refused or failed on; their tokens count for no row, whether the row refused or answered them, and a refused item still counts as a wrong answer for Intelligence. This row: USD 0.051514 per 1,000 decisions (0.797x the official cost cap USD 0.064595), against USD 0.050999 over all 1,500 items (v1.6.29).
Measured p50 latency: 0.035 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: 0.22 s (est. incl. network (2x + 150 ms))
Self-hosted model: measured p50 0.035 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 = 0.22 s. Model authors can ask us to time a hosted endpoint instead (on request).
x2 + 0.15 s (assumption, not measured)
evaluator-owned Lium GPU pod (H100), offline read-only container (--network none); author's reference server code/serve_head.py, Quick-start argv, unchanged defaults Paid fast-lane order 3a081599 (public). Same customer and reviewed route as standard_one_8b_sh: official pytorch/pytorch:2.11.0-cuda12.8-cudnn9-runtime plus the documented requirements (transformers 5.12.1, peft 0.21.0) installed offline from SHA-256-pinned PyPI wheels (PR #299, bound to this order by PR #332). Requests over the documented 32,768-token context answer HTTP 422 (refused, not truncated). Customer list price USD 0.038/M input is below the M2 base floor.