Qwen 3.8 Max Prime

Alibaba, closed weights, released Sep 2026, run at provider defaults through OpenRouter, served by Alibaba. It built the same five pages as every model on season 0.6.1 of the Rams Index, and Rams graded each one at desktop and phone size.

Rams Index, 0 to 100

7.5

95% interval 7.1–11.0

Rank: 24of 26 models by Index; the paired intervals allow 22 to 24
Cost per page: $0.34Generation only, list price on run day
Earlier judges: 84.8Under the grade sum this board used until 10/28, GPT-5.6 Terra placed it fourth of 26 and the Claude judge first.
Grade sum: 29.08Seven 1–5 grades, out of 35
Pages judged: 1010 judge passes, 2 builds per brief
Defects per page: 14.7Text under 12px (10 of 10)

Rank 24 of 26; the data allows anywhere from 22 to 24. Within noise of MiniMax M3 and Qwen 3.8 Max: the gap to each sits inside the noise, so the board can't order them yet. Clearly behind GPT-6.1 Sol, GPT-6 Astra, Grok 4.7, Claude Fable 5.1, GPT-6.1 Terra, GLM 5.3, Claude Opus 5.5, Claude Sonnet 5.5, Mistral Large 4, GPT-6 Luna, Claude Haiku 5.5, GPT-5.6 Sol, Claude Fable 5, Kimi K3, Kimi K3.1 Pro, GLM 5.3 Prime, DeepSeek V4 Pro, Claude Opus 4.8, MiMo V2.6 Pro, Gemini 4 Argon and Gemini 3.8 Flash. Clearly ahead of Muse Spark Open and Muse Wonder.

Checked against one designer · 63% agreementSeason 0.6.1New this seasonProvisional: 1.0 judge passes per page, protocol 2

Grades by dimension

The seven grades behind the Index, 1 to 5, mean of desktop and phone. The tick is the mean across all 26 models.

Hierarchy4.251 of 26
Typography3.952 of 26
Spacing3.351 of 26
Color4.002 of 26
Components3.756 of 26
Composure3.851 of 26
Craft3.902 of 26

Bars start at 2.8 and end at 4.6, the range the board uses, so small differences show.

Index by brief

The same Index on each of the five briefs, and where the model places on that brief among all 26.

Landing30.817 of 26
Dashboard21.618 of 26
Settings1.925 of 26
Pricing1.825 of 26
Article7.523 of 26

Bars run from 0 to 100.

How it ran

Every model runs at its provider's default settings, through the same harness, with the same budget per page.

Effort: DefaultThe provider’s default. We send no effort or thinking setting
Time per page: 303sMedian over every build, broken ones included
Output tokens: 26.0kMedian per page, reasoning included
Reasoning tokens: 8.9kMedian per page, as the API reports them
Broken pages: 0 of 10Every build came back as a whole page
Criticals: 50%5 of 10 scored pages had a critical issue in their first Rams code review
Built through: OpenRouterServed by Alibaba

Best and worst page

Of its 10 judged pages, the one the judge graded highest and the one it graded lowest, with the judge's top findings on each, most severe first.

Best: Landing37.7
Index on this page, one judge pass. Brief mean 30.8.
- Redundant content between ticker and hero card badgesModerate
- Subline paragraph sits unusually far below headlineModerate
- Invoice badge '#214' label and bold span visually gap mid-stringModerate
Worst: Article1.3
Index on this page, one judge pass. Brief mean 7.5.
- Corrupted kicker textSerious
- Kicker text rendering corruptionSerious
- Reading-progress counter shows '0' at article top with no in-motion stateModerate