THE MODEL GAUNTLET

Capabilities.
Put to the test.

Original development challenges. Measured technical evidence. Independent human judgement.

60% technical assessment40% human reviewCapped by the reviewer’s quality bandHow we score ↗

OVERALL STANDINGS / METHODOLOGY 1.1

Every challenge, judged.

Updated 23 October 2026

Ranked by the B01–B05 total out of 500 (methodology 1.1). Wildcards are scored separately, out of 100 each.
Rank / model	B01	B02	B03	B04	B05	Total /500	B06	B07
1Opus 5.5 ↗	98.2	95.2	92.4	90.3	88.2	464.2	100	100
2Astra ↗	90.5	95	89.6	80	80	435.1	95	80
3MiniMax M3.1 Flash ↗	91.8	90	91.8	71.3	80	424.9	90	60
4Beam 1.0 501B A23B ↗	80	50	60	59.8	40	289.8	40	10

THE RENDERS / BUILT IN BLENDER

B02 / 4 MODELS

B06 / 4 MODELS

B03 / CHALLENGE RESULTS

A playable Godot environment that carries a Blender asset through to a working game.

Final = 0.6 × technical + 0.4 × human, capped by the band the reviewer chose.
Rank / model	Technical /100	Human /100	Final /100
1Opus 5.5 ↗	90Ceiling 100	96Playability · Good	92.4Not capped
2MiniMax M3.1 Flash ↗	97Ceiling 100	84Playability · Good	91.8Not capped
3Astra ↗	88Ceiling 100	92Playability · Good	89.6Not capped
4Beam 1.0 501B A23B ↗	81Ceiling 100	52Playability · Weak	60Capped at 60 by the band

COMPARE THE EVIDENCE

Side by side, test by test, under the same brief and methodology.