Benchmarks · Writing & design
GameFeel Benchmark v1.5.2
Star0The GameFeel Benchmark: each model adds 30 to 100 items, mechanics and balance changes of its own choosing to 12 unpublished game projects. Score: the board's Overall, the mean of its seven subcategories (creating, bug fixing, gamefeel, ideas, visuals, audio and reverse evaluation), out of 100. Gamefeel itself, how well the additions keep each game balanced, is rated 1 to 10 per task by human evaluators. All seven subcategory scores are kept in each row's details.
Quality 71 of 100 · Private test set · Has measured 6 of today's 10 default lineup models
23 models · 30 results · updated 2026-11-23 · kodasdream.github.io ↗
Leaderboard
Benchmark quality
How the score works- Overall71
71 of 100
How useful it is for telling today’s models apart
- Headroom20%49
Top models average 46 of 100
Can top models still improve on it?
- Test set20%100
Private
Could labs train on the questions?
- Recency15%100
3 days old
How new is it?
- Grading15%70
Partly automatic
How are answers scored?
- New models15%60
6 of 10 top models
How fast does it test new releases?
- Efforts10%35
4 of 23 models at 2+ efforts
Does it test each model at several reasoning efforts?
- Breadth5%57
23 models
How many models has it measured?
How each part was worked out
- Headroom 49 of 100 · 20% · +9.8
- From the spread of published scores and the room left above the top quarter of models, one score per model.
- Test set 100 of 100 · 20% · +20
- Private test set. The 12 game projects are unpublished, so the scored content is hidden; only the CSV of scores is public. Source
- Recency 100 of 100 · 15% · +15
- Released 2026-11-19 (from the publisher's pages, not yet checked by hand).
- Grading 70 of 100 · 15% · +10.5
- Coding answers are checked against answer keys (Opus 5.5 only rules on formatting slips and near misses, never on quality); game feel, ideas, visuals and audio are rated by people.
- New models 60 of 100 · 15% · +9
- Has measured 6 of today's 10 default lineup models. Too few new models to time how quickly it adds them.
- Efforts 35 of 100 · 10% · +3.5
- Measured 4 of the 23 models that come in more than one reasoning effort at two or more of them.
- Breadth 57 of 100 · 5% · +2.9
- 23 models measured.
Not counted in the score: 0 stars · used by 2 published pages. Both lift it in the directory’s recommended order. Test set, release date and grading were researched from the publisher’s pages (2026-11-19); tell us if one is wrong.
More from this benchmark
9 published measuresScore vs any published measure
13 more models under All models in settings.
Breakdown
darker is better within each column · click a column to sort| Model | ||||||||
|---|---|---|---|---|---|---|---|---|
| Fable 5.2 max | 71.0 | 69.0 | 51.0 | 42.0 | 64.0 | 31.0 | 18.0 | 102.0 |
| Gemini 4 Carbon high | 64.0 | 56.0 | 51.0 | 47.0 | 67.0 | 33.0 | 23.3 | 48.0 |
| Gemini 4 Neon xhigh | 63.0 | 74.0 | 48.0 | 42.0 | 55.0 | 39.0 | 20.0 | 37.0 |
| Gemini 4 Hydrogen high | 63.0 | 59.0 | 54.0 | 50.0 | 65.0 | 32.0 | 11.8 | 24.0 |
| Opus 5.5 xhigh | 65.0 | 65.0 | 50.0 | 41.0 | 63.0 | 25.0 | 17.9 | 90.0 |
| Gemini 4 Argon high | 63.0 | 55.0 | 35.0 | 30.0 | 65.0 | 30.0 | 25.4 | 35.0 |
| ERNIE 5.5 Flash Preview max | 64.0 | 55.0 | 45.0 | 36.0 | 56.0 | 44.0 | 3.07 | |
| Sonnet 5.5 max | 65.0 | 65.0 | 48.0 | 39.0 | 54.0 | 21.0 | 8.21 | 90.0 |
| DeepSeek V4.1 Flash 1017 max | 65.0 | 54.0 | 37.0 | 32.0 | 55.0 | 50.0 | 5.34 | 85.0 |
| MiMo V2.6 Pro max | 52.0 | 65.0 | 35.0 | 38.0 | 47.0 | 41.0 | 1.93 | 80.0 |
| DeepSeek V4.1 Flash max | 62.0 | 57.0 | 45.0 | 34.0 | 35.0 | 43.0 | 1.56 | 45.0 |
| Qwen 3.8 Flash xhigh | 55.0 | 55.0 | 50.0 | 42.0 | 27.0 | 40.0 | 4.03 | 15.0 |
| Qwen 3.8 Max xhigh | 56.0 | 40.0 | 47.0 | 41.0 | 30.0 | 42.0 | 5.7 | 15.0 |
| Haiku 5.5 max | 56.0 | 50.0 | 43.0 | 31.0 | 48.0 | 19.0 | 1.3 | 22.0 |
| Gemini 3.8 Flash max | 55.0 | 55.0 | 20.0 | 30.0 | 37.0 | 39.0 | 0.854 | 75.0 |
| Ling 3.2 Flash max | 40.0 | 45.0 | 30.0 | 30.0 | 40.0 | 32.0 | 1.08 | 48.0 |
| GLM 5.3 Flash max | 57.0 | 62.0 | 32.0 | 25.0 | 18.0 | 22.0 | 0.331 | 270.0 |
| Kimi K3.1 max | 63.0 | 55.0 | 10.0 | 10.0 | 45.0 | 27.0 | 3.53 | 77.0 |
| Qwen 3.7 Plus medium | 35.0 | 30.0 | 30.0 | 30.0 | 20.0 | 20.0 | 3.68 | 15.0 |
| Muse Spark 1.3 medium | 33.0 | 50.0 | 15.0 | 15.0 | 25.0 | 25.0 | 0.358 | 90.0 |
| MiniMax M3.1 max | 20.0 | 63.0 | 0 | 10.0 | 30.0 | 30.0 | 0.395 | 135.0 |
| GPT-6 Luna xhigh | 50.0 | 15.0 | 0 | 15.0 | 28.0 | 15.0 | 0.0259 | 96.0 |
| Nemotron 3.5 Super VL max | 9 | 2 | 27.0 | 25.0 | 25.0 | 25.0 | 3.57 | 9 |
All results
30 rows| # | Model | Effort | Score |
|---|---|---|---|
| 1 | Fable 5.2 | max | 47.3 |
| 2 | Gemini 4 Carbon | high | 47.1 |
| 3 | Gemini 4 Neon | xhigh | 46.8 |
| 4 | Fable 5.2 | medium | 46.7 |
| 5 | Fable 5.2 | high | 46.7 |
| 6 | Gemini 4 Hydrogen | high | 45.4 |
| 7 | Opus 5.5 | xhigh | 44.8 |
| 8 | Opus 5.5 | max | 44.0 |
| 9 | Gemini 4 Neon | medium | 43.8 |
| 10 | Gemini 4 Argon | high | 42.1 |
| 11 | ERNIE 5.5 Flash Preview | max | 40.6 |
| 12 | Sonnet 5.5 | max | 40.6 |
| 13 | DeepSeek V4.1 Flash 1017 | max | 40.1 |
| 14 | Sonnet 5.5 | xhigh | 39.3 |
| 15 | Opus 5.5 | medium | 38.9 |
| 16 | MiMo V2.6 Pro | max | 37.5 |
| 17 | DeepSeek V4.1 Flash | max | 37.1 |
| 18 | Qwen 3.8 Flash | xhigh | 36.7 |
| 19 | Sonnet 5.5 | medium | 35.8 |
| 20 | Qwen 3.8 Max | xhigh | 35.3 |
| 21 | Haiku 5.5 | max | 33.2 |
| 22 | Gemini 3.8 Flash | max | 31.6 |
| 23 | Ling 3.2 Flash | max | 29.1 |
| 24 | GLM 5.3 Flash | max | 28.9 |
| 25 | Kimi K3.1 | max | 28.7 |
| 26 | Qwen 3.7 Plus | medium | 22.7 |
| 27 | Muse Spark 1.3 | medium | 21.8 |
| 28 | MiniMax M3.1 | max | 20.5 |
| 29 | GPT-6 Luna | xhigh | 16.4 |
| 30 | Nemotron 3.5 Super VL | max | 15.8 |