ENPT

Amos 1

Last updated:

Open27–122ms$0/M input

Quadrant scores

See the full quadrant

Scored with the public quadrant rubric (maturity × capability; bubble = adoption). Revise as evidence lands.

  • Maturity5.9/10

    Four Apache-2.0 open-weight sizes (large 27B on Qwen3.8-27B, medium 9B, small 2B, mini 0.5B on an Ettin encoder + SigLIP2) plus a free hosted API at amos.scira.ai. Detailed cards. Solo maintainer, no SLA.

  • Capability8.3/10

    Official boards (17 Oct): Decision Index 0.3.2 Full 64.08 (#6); JevBench Intelligence 69.7, calibration 89.0, p50 0.51 s. Choice / yes-no / score, images, and boxes on named objects. 131,072 tokens per request.

  • Adoption2/100

    Launch thread 522 likes / 19K views (13 Oct). HF repos are hours old (single-digit downloads). No integrations yet.

Vendor claims

  • Amos 1 large: Decision Index 0.3 public part 65.11 (board kit, own run); Decision Bench 94.9% on 1,071 records (Perplexity Decider v1.1 94.5%, Jev 1.13 93.8% on that leaderboard); JevBench public hard 84.7[Vendor claim — not independently verified]
  • Median server time per three-question request: large 122 ms, medium 61 ms, small 48 ms, mini 27 ms (93 ms on a laptop CPU); small 2B places boxes on the named object in 97.5% of COCO photos[Vendor claim — not independently verified]

News covering this model

publicado (OK do Fabricio 17/10 BRT). Board audit of 17 Oct 2026: see the “Board audit” section below; where it conflicts with older board statements on this page, it wins.

Read this first. All numbers are the author’s (UnverifiedClaim). The 65.11 is the public part of Decision Index 0.3 run with the board’s own kit; on the board’s public column it would be #3 (behind 65.64 and 65.45). The full board score adds private tests only the maintainers run.

Amos 1 (13 Oct 2026, by Zaid Mukaddam of Scira) is four open decision models under Apache-2.0: large 27B (on Qwen3.8-27B), medium 9B, small 2B and mini 0.5B (an Ettin 400M encoder with a SigLIP2 vision tower). You send a case and typed questions and get a probability for every answer in one pass. It reads photos, and for “where is…?” questions it returns a box.

Hosted API, free for now: amos.scira.ai.

Author results (UnverifiedClaim)

Amos 1 large
Decision Index 0.3, public part (board kit) 65.11
Decision Bench (1,071 records) 94.9%
JevBench public easy / standard / hard 100.0 / 100.0 / 84.7
RefCOCO box centred on named object 91.9%
typed-decisions accuracy / Brier 76.1% / 0.069

On typed-decisions the card itself shows Microsoft-Decision-1 ahead on accuracy (79.5%) and Decider 1 slightly ahead (76.8%); Amos’s edge there is calibration.

Score working (13 Oct 2026)

  • Maturity = 9 × 0.30 + 7 × 0.25 + 4 × 0.25 + 2 × 0.20 = 2.70 + 1.75 + 1.00 + 0.40 = 5.85 → 5.9
  • Capability = 10 × 0.30 + 7 × 0.25 + 7 × 0.25 + 7 × 0.20 = 3.00 + 1.75 + 1.75 + 1.40 = 7.90 → 7.9
  • Adoption = 6 × 0.35 + 1 × 0.35 + 0 × 0.30 = 2.10 + 0.35 + 0 = 2.45 → 2

Types 10: all three types plus images and boxes. Accuracy 7: own run with the board’s kit, counted as evidence per Fabricio’s rule, not on the board yet.

Board audit (17 Oct 2026)

publicado (OK do Fabricio 17/10 BRT). Checked against the official boards on 17 Oct 2026: Decision Index 0.3.2 (data of 16 Oct, 19:00 UTC), JevBench and ImageJevBench live. The boards are separate sources and are never averaged. Older board statements and rankings further down this page are superseded where they conflict with this section.

Board Row Result
Decision Index 0.3.2 (16 Oct 19:00 UTC) Amos 1 Large (27B) Full 64.08 (#6, tie group 3); public 64.94 (#7); same-skills 62.69; new-domains 55.53
JevBench API (live 17 Oct) Amos 1 large Capability 79.4; Intelligence 69.7; calibration 89.0; p50 0.51 s; $0.54/1k
JevBench API (live 17 Oct) Amos 1 medium Capability 63.2; Intelligence 38.4; calibration 88.0; p50 0.44 s; $0.13/1k

Board-status statements on this page to recheck against the table (may be superseded):

  • “Accuracy 7: own run with the board’s kit, counted as evidence per Fabricio’s rule, not on the board yet.”

Score update (audit, 17 Oct 2026). Capability 7.9 → 8.3 = types × 0.30 + calibration × 0.25 + latency × 0.25 + accuracy × 0.20 = 10 × 0.30 + 7 × 0.25 + 7 × 0.25 + 9 × 0.20 = 8.30 → 8.3. Types, calibration and latency come from the earlier working (10 / 7 / 7); only accuracy is re-derived from the board (9; was 7): Decision Index Full 64.08 and JevBench Intelligence 69.7 are both above the Jev 1.13 references (60.11 and 63.6), two independent sources, so 9. Maturity and adoption unchanged.

Hugging Face on 17 Oct (mukaddamzaid/amos-1-large): 116 downloads, 16 likes, licence tag apache-2.0, last modified 2026-10-16.