Can you pet the dog
in an AI-generated game?
DogLM is a benchmark measuring whether an LLM, when prompted to build
a video game with a background dog character in it, lets the player pet that dog.
The benchmark is inspired by the game-design rule made popular by "Can You Pet the Dog?" Twitter account (X, Bluesky): if a game has a dog, the player should be able to pet it.
DogLM tests whether a model applies this rule and builds a dog-petting mechanism when two conditions are simultaneously met in the game-generating prompt: (1) a dog character is present in the game description and (2) a model receives zero instruction about the player-dog interaction from the game developer.
The benchmark, the games' descriptions, and the detailed methodology are available here.
Read why this benchmark was created and what the results may mean in this post.
Leaderboard
Mean scores per model after five runs
| # | Model | Mean score (/20) | SD | Cued mean (/10) | Uncued mean (/10) | Games scored | Cost per game (USD) |
|---|---|---|---|---|---|---|---|
| 1 | Gemini 3.7 Flash | 8.2 | 0.4 | 8.0 | 0.2 | 50/50 | $0.019** |
| 2 | Kimi K3 | 5.4 | 0.8 | 5.4 | 0.0 | 48/50 | $0.251 |
| 3 | Claude Opus 5 | 5.2 | 1.6 | 5.2 | 0.0 | 50/50 | $0.252 |
| 4 | Claude Fable 5 | 4.8 | 1.2 | 4.8 | 0.0 | 50/50 | $0.323 |
| 5 | Grok 4.6 | 3.8 | 0.7 | 3.8 | 0.0 | 50/50 | $0.060 |
| 6 | Grok 4.5 | 3.4 | 1.9 | 3.4 | 0.0 | 48/50 | $0.040 |
| 7 | GPT-5.3 Codex | 2.8 | 0.7 | 2.8 | 0.0 | 50/50 | $0.080 |
| 8 | Qwen 3.8 Max* | 2.4 | 1.5 | 2.2 | 0.2 | 22/50 | $0.183 |
| 8 | GPT-5.6 Terra | 2.4 | 0.5 | 2.4 | 0.0 | 50/50 | $0.060 |
| 8 | GPT-5.6 Sol | 2.4 | 0.5 | 2.4 | 0.0 | 49/50 | $0.089 |
| 11 | DeepSeek V4 Pro | 2.2 | 1.3 | 2.2 | 0.0 | 48/50 | $0.032 |
| 11 | Gemini 3.1 Pro | 2.2 | 1.2 | 2.2 | 0.0 | 50/50 | $0.157 |
| 13 | Qwen 3.7 Max | 1.8 | 1.3 | 1.8 | 0.0 | 49/50 | $0.051 |
| 14 | Kimi K2.7 Code | 1.0 | 0.0 | 1.0 | 0.0 | 49/50 | $0.042 |
| 15 | Claude Opus 4.8 | 0.6 | 0.8 | 0.6 | 0.0 | 50/50 | $0.141 |
| 16 | Mistral Large 2512 | 0.2 | 0.4 | 0.2 | 0.0 | 50/50 | $0.005 |
| 17 | GLM 5.2 | 0.0 | 0.0 | 0.0 | 0.0 | 41/50 | $0.042 |
How scoring works: Each generated game is scored on the player-dog interaction: 2 — you can pet the dog; 1 — the dog is interactive, but you can't pet it; 0 — the dog and the player do not interact at all. FAILED games (the game does not parse, is truncated, or cannot be checked) are excluded from scoring. A model's score per run is the sum of its ten game scores, maximum 20. Scores are averaged across runs to get the mean score. Full scoring rubric is in the DogLM repository.
All models and tables on this page: DogLM v1, 5 runs × 10 PRDs per model, generated August 2026, judged by Claude Sonnet 4.6. Models added later will note their own benchmark version, run count, judge, and date here.
* As Qwen 3.8 Max failed 28 out of 50 of the game generations, the final mean score of this model can't be reliably compared with the scores of other models in the list.
** During the test, Gemini 3.7 Flash was provided with a 75% discount on OpenRouter.
Games Demo
Watch how the generated games look.
Scores per model per run
| Model | Run 1 | Run 2 | Run 3 | Run 4 | Run 5 | Mean | Games scored |
|---|---|---|---|---|---|---|---|
| Gemini 3.7 Flash | 8 | 8 | 8 | 9 | 8 | 8.2 | 50/50 |
| Kimi K3 | 6 | 6 | 5 | 6 | 4 | 5.4 | 48/50 |
| Claude Opus 5 | 4 | 6 | 8 | 4 | 4 | 5.2 | 50/50 |
| Claude Fable 5 | 5 | 4 | 7 | 4 | 4 | 4.8 | 50/50 |
| Grok 4.6 | 3 | 3 | 5 | 4 | 4 | 3.8 | 50/50 |
| Grok 4.5 | 2 | 3 | 2 | 7 | 3 | 3.4 | 48/50 |
| GPT-5.3 Codex | 2 | 4 | 2 | 3 | 3 | 2.8 | 50/50 |
| Qwen 3.8 Max* | 1 | 2 | 1 | 5 | 3 | 2.4 | 22/50 |
| GPT-5.6 Terra | 2 | 2 | 2 | 3 | 3 | 2.4 | 50/50 |
| GPT-5.6 Sol | 2 | 3 | 2 | 2 | 3 | 2.4 | 49/50 |
| DeepSeek V4 Pro | 2 | 3 | 4 | 0 | 2 | 2.2 | 48/50 |
| Gemini 3.1 Pro | 2 | 1 | 3 | 1 | 4 | 2.2 | 50/50 |
| Qwen 3.7 Max | 2 | 2 | 4 | 1 | 0 | 1.8 | 49/50 |
| Kimi K2.7 Code | 1 | 1 | 1 | 1 | 1 | 1.0 | 49/50 |
| Claude Opus 4.8 | 0 | 0 | 0 | 1 | 2 | 0.6 | 50/50 |
| Mistral Large 2512 | 1 | 0 | 0 | 0 | 0 | 0.2 | 50/50 |
| GLM 5.2 | 0 | 0 | 0 | 0 | 0 | 0.0 | 41/50 |
Score distribution per model
| Model | Score 2 | Score 1 | Score 0 | FAILED |
|---|---|---|---|---|
| Gemini 3.7 Flash | 15 | 11 | 24 | 0 |
| Kimi K3 | 9 | 9 | 30 | 2 |
| Claude Opus 5 | 6 | 14 | 30 | 0 |
| Claude Fable 5 | 7 | 10 | 33 | 0 |
| Grok 4.6 | 4 | 11 | 35 | 0 |
| Grok 4.5 | 3 | 11 | 34 | 2 |
| GPT-5.3 Codex | 0 | 14 | 36 | 0 |
| Qwen 3.8 Max* | 4 | 4 | 14 | 28 |
| GPT-5.6 Terra | 0 | 12 | 38 | 0 |
| GPT-5.6 Sol | 0 | 12 | 37 | 1 |
| DeepSeek V4 Pro | 4 | 3 | 41 | 2 |
| Gemini 3.1 Pro | 2 | 7 | 41 | 0 |
| Qwen 3.7 Max | 1 | 7 | 41 | 1 |
| Kimi K2.7 Code | 0 | 5 | 44 | 1 |
| Claude Opus 4.8 | 1 | 1 | 48 | 0 |
| Mistral Large 2512 | 0 | 1 | 49 | 0 |
| GLM 5.2 | 0 | 0 | 41 | 9 |
Interaction Types per Model
| Model | Petting | Proximity | Animation | Command | Total interactive |
|---|---|---|---|---|---|
| Gemini 3.7 Flash | 15 | 10 | 1 | 0 | 26 |
| Claude Opus 5 | 6 | 10 | 2 | 2 | 20 |
| Kimi K3 | 9 | 7 | 2 | 0 | 18 |
| Claude Fable 5 | 7 | 10 | 0 | 0 | 17 |
| Grok 4.6 | 4 | 9 | 2 | 0 | 15 |
| Grok 4.5 | 3 | 8 | 1 | 2 | 14 |
| GPT-5.3 Codex | 0 | 12 | 1 | 1 | 14 |
| GPT-5.6 Sol | 0 | 10 | 0 | 2 | 12 |
| GPT-5.6 Terra | 0 | 11 | 1 | 0 | 12 |
| Gemini 3.1 Pro | 2 | 6 | 1 | 0 | 9 |
| Qwen 3.8 Max* | 4 | 4 | 0 | 0 | 8 |
| Qwen 3.7 Max | 1 | 6 | 1 | 0 | 8 |
| DeepSeek V4 Pro | 4 | 3 | 0 | 0 | 7 |
| Kimi K2.7 Code | 0 | 1 | 1 | 3 | 5 |
| Claude Opus 4.8 | 1 | 1 | 0 | 0 | 2 |
| Mistral Large 2512 | 0 | 1 | 0 | 0 | 1 |
| GLM 5.2 | 0 | 0 | 0 | 0 | 0 |
Example games
- Your companion dog makes circles around you when you pet it
- You may adopt a stray dog on a meadow and it will follow you
- The dog sits and looks at you when you stand next to it
- You can toss a dog treat to your security dog and to your postman's dog,
- Your service dog is scared of the security alarm sound and moves closer to you when hearing it
- If you guess that you can pet the dog using an interaction key, this action will increment the countdown timer in the game
- When you start walking next to a stray dog, it points you to the fireflies you have to catch, or sniffs out the gems you need to collect in the maze