Benchmarks · Coding · Agents & computer use
KingBench 4
Star0Six hard apps, each built from one prompt in the model's own coding tool within 60 minutes: a Game Boy, DS and 3DS emulator, a Docker alternative, a 3D room designer, a QMK keyboard configurator, Pokémon Red and a Fitbit desktop app. AICodeKing grades each out of 10 as delivered; the total out of 60, as a percent. One person's grading, and a model plus its tool.
Quality 51 of 100 · Public test set · Has measured 9 of today's 10 default lineup models
16 models · 16 results · updated 2026-11-18 · youtube.com ↗
Leaderboard
Benchmark quality
How the score works- Overall51
51 of 100
How useful it is for telling today’s models apart
- Headroom20%36
Top models average 62 of 100
Can top models still improve on it?
- Test set20%30
Public
Could labs train on the questions?
- Recency15%100
43 days old
How new is it?
- Grading15%50
One person judges
How are answers scored?
- New models15%90
9 of 10 top models
How fast does it test new releases?
- Efforts10%0
One effort per model
Does it test each model at several reasoning efforts?
- Breadth5%40
16 models
How many models has it measured?
How each part was worked out
- Headroom 36 of 100 · 20% · +7.2
- From the spread of published scores and the room left above the top quarter of models, one score per model.
- Test set 30 of 100 · 20% · +6
- Public test set. The six build prompts are shown in AICodeKing's launch video; the author runs each model in its coding tool and grades every app himself, out of 10. Source
- Recency 100 of 100 · 15% · +15
- Released 2026-10-10 (from the publisher's pages, not yet checked by hand).
- Grading 50 of 100 · 15% · +7.5
- Graded by hand by one person, the benchmark's author.
- New models 90 of 100 · 15% · +13.5
- Has measured 9 of today's 10 default lineup models. Too few new models to time how quickly it adds them.
- Efforts 0 of 100 · 10% · 0
- Measured 0 of the 16 models that come in more than one reasoning effort at two or more of them.
- Breadth 40 of 100 · 5% · +2
- 16 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
1 published measuresScore vs any published measure
5 more models under All models in settings.
All results
16 rows| # | Model | Effort | KingBench 4 |
|---|---|---|---|
| 1 | Fable 5.2 | high | 70.0% |
| 2 | Gemini 4 Neon | high | 63.2% |
| 3 | Opus 5.5 | high | 57.8% |
| 4 | GPT-6 Astra | high | 56.5% |
| 5 | GPT-6.2 Astra | high | 56.0% |
| 6 | GPT-6.1 Sol | high | 55.0% |
| 7 | Gemini 4 Argon | high | 54.3% |
| 8 | Haiku 5.5 | high | 54.2% |
| 9 | Kimi K3.1 | high | 53.7% |
| 10 | Gemini 4 Carbon | high | 53.3% |
| 11 | Muse Wonder Full | high | 53.0% |
| 12 | Qwen 4 Max | high | 51.0% |
| 13 | GPT-6.2 Sol | high | 49.3% |
| 14 | Muse Wonder | high | 45.8% |
| 15 | GLM 5.3 | high | 45.8% |
| 16 | Gemini 4 Hydrogen | high | 41.5% |