glyph-cluster / README.md
stealth-model's picture
Update card: StealthMark 1.8.10, plainer links, softer panel backgrounds
3380c70 verified
|
Raw History Blame Contribute Delete
2.37 kB
metadata
license: cc-by-4.0
tags:
  - glyph-cluster
  - stealth-model
  - stealthmark
  - coding
  - reasoning
  - long-context
  - tool-use
thumbnail: https://stealthmodels.com/glyph-cluster/assets/social-card-v2.png
Glyph Cluster: four colorful glyph characters fitting together like a block puzzle, with its 256K context.

Glyph Cluster counts tokens exactly like Microsoft's MAI Thinking 1, on all 24 of our test phrases. That's our lead for who made it.

By Stealth Models.

StealthMark 1.8.10 results for Glyph Cluster: benchmark score 18.9, General Intelligence 12.0, Visual Intelligence 29.8. Strongest at composing a whole scene: 46.3 on Beach. Every StealthMark result of Glyph Cluster: Beach 46.3, Raccoon 35.7, Selfie 32.4, Ground Truth 19.9, Newton's Mirror 12.8, Mona Lisa 11.7, Foresight 6.4, Latch 3.6.

Compare it with MAI Thinking 1

The case file. First seen 7 Oct 2026, revealed not yet, developer unknown, context 256K tokens, input Text only, tool calls Yes (function calling), speed in our tests about 69 tok/s. 9 Oct 2026, our finding: Our lead: Microsoft’s MAI. Its token counts and its Newton’s Mirror failures match MAI Thinking 1. 25 Oct 2026, our finding: Iceman Contributor counts tokens exactly like Glyph Cluster: one tokenizer family behind both.

See our verdict

Facts as text
Fact Value
First seen 7 Oct 2026
Revealed not yet
Developer unknown
Context 256K tokens
Input Text only
Tool calls Yes (function calling)
Speed in our tests about 69 tok/s
Access Active as a free preview; prompts may be retained for training

StealthMark leaderboard
Text and images: CC BY 4.0, stealthmodels.com.