Abstract
Count of superlatives in the first public announcement of a model: "smartest", "best", "world's top", "state-of-the-art", "frontier", "AGI", and "a GPT-3 to GPT-4 moment". Unreleased models are eligible, since their launch posts are often the only artifact available.
Results
Superlatives · count · Lower is better
Methodology
- Each superlative is counted once per sentence. "The smartest AI in the world by a significant margin on every metric" counts as three.
- Fan-account amplification is excluded, or this chart would need a log scale.
Limitations
- We count claims, not capability. The second number has been more stable.
Results table
| # | Model | Superlatives |
|---|---|---|
| 1 | Claude Nextclaimed | 0 |
| 2 | Claude Sonnet 5.5 | 3 |
| 3 | Claude Opus 5.5 | 4 |
| 4 | DeepSeek V4.1 Flash | 5 |
| 5 | Kimi K3 | 6 |
| 6 | GPT-6 Luna | 6 |
| 7 | Claude Haiku 5.5 | 6 |
| 8 | Claude Fable 5.1 | 6 |
| 9 | GPT-6.1 Sol | 7 |
| 10 | GLM-5.3 | 7 |
| 11 | GPT-6 Astra | 8 |
| 12 | Qwen3.8 Max | 9 |
| 13 | Reflection 405Bclaimed | 9 |
| 14 | Mistral Large 4 | 10 |
| 15 | Gemini 4 Argonclaimed | 11 |
| 16 | Le Chaton Fatclaimed | 12 |
| 17 | Muse Wonder | 12 |
| 18 | Gemini 3.8 Flash | 12 |
| 19 | Muse Wonder Full | 14 |
| 20 | Grok 5claimed | 14 |
| 21 | Grok 4.7 | 17 |
| 22 | Our Model™claimed | 31 |
Citation
@misc{tmb2026superlatives,
title = {Superlatives per Launch Post: "Smartest", "best in the world", "AGI", per announcement},
author = {Bro, Trust Me and {et al.}},
year = {2026},
howpublished = {\url{https://trustmebrobenchmarks.com/benchmarks/superlatives/}},
note = {Results not reproducible. Please do not cite.}
}