Abstract
Models write a tic-tac-toe opponent in Brainfuck. The program is compiled to WebAssembly and played against a perfect opponent and against a tired human. Tiers run from S+ (never loses, occasionally gloats) to F (does not compile).
Results
Brainfuck · tier · Higher is better
Methodology
- Programs exceeding 1 MB of '+' characters are disqualified for unsportsmanlike conduct.
- Tier boundaries: S+ ≥ 95, S ≥ 85, A ≥ 70, B ≥ 55, C ≥ 40, D ≥ 25, F below that.
Limitations
- Brainfuck competence may not transfer to your codebase. Then again, have you seen your codebase.
Results table
| # | Model | Brainfuck |
|---|---|---|
| 1 | Claude Opus 5.5 | S |
| 2 | GPT-6.1 Sol | A |
| 3 | Claude Fable 5.1 | A |
| 4 | Claude Sonnet 5.5 | A |
| 5 | GPT-6 Astra | B |
| 6 | GPT-6 Luna | B |
| 7 | DeepSeek V4.1 Flash | B |
| 8 | GLM-5.3 | B |
| 9 | Muse Wonder Full | C |
| 10 | Grok 4.7 | C |
| 11 | Gemini 3.8 Flash | C |
| 12 | Kimi K3 | C |
| 13 | Qwen3.8 Max | D |
| 14 | Muse Wonder | D |
| 15 | Claude Haiku 5.5 | D |
| 16 | Mistral Large 4 | D |
Citation
@misc{tmb2026brainfuck,
title = {Brainfuck Tic-Tac-Toe Tier: Eight characters. Nine squares. Zero excuses.},
author = {Bro, Trust Me and {et al.}},
year = {2026},
howpublished = {\url{https://trustmebrobenchmarks.com/benchmarks/brainfuck/}},
note = {Results not reproducible. Please do not cite.}
}