Game development · Models
Design Arena's Elo for game development · models, from pairwise human preference votes, as OpenRouter carries it. 120 models ranked, as carried in OpenRouter's catalog.
retrieved sha256 · 0bb53a77713fwww.designarena.ai ↗
7 models sit on the frontier: nothing scores higher for less.
- the attractive quadrant
- the frontier
- a priced model, in its maker's tone
- on the frontier, named
- 01Hy4 flash1225 · $0.061 blended
- 02DeepSeek V4.1 Flash 10171253 · $0.152 blended · +28 at 2.5x
- 03Claude Haiku 5.51313 · $0.20 blended · +60 at 1.3x
- 04Muse Spark 1.31344 · $2.00 blended · +31 at 10x
- 05Kimi K3.1 Pro1402 · $2.281 blended · +58 at 1.1x
- 06GPT-6.1 Sol1406 · $4.00 blended · +4 at 1.8x
- 07Claude Opus 5.51427 · $8.00 blended · +21 at 2.0x
120 of 120 ranked models carry a list price; 0 are free or unlisted and not in the field. Price is the blended OpenRouter rate, three input tokens to one output, on a logarithmic axis. Each dot is filled in its maker's tone, as the columns above are. The numbered and named dots are the frontier, cheapest first; each step after the first names what it adds over the step before, and at what multiple of the blended price. A hover over any dot names it and its two figures, a click opens the model, and a model named in the picker above keeps its label on. 17 of 120 priced models sit in the attractive quadrant: at or above the median Elo of 1183 and at or below the median blended price of $1.026. The tinted corner is that quadrant. 18 of the 120 crosses match the selection below and are drawn white.
18 of 120.
Design Arena · Elo7 more authors are on this board, the largest with 3 models. Search a name to reach one.
| Position | Model | Author | Elo | Win rate | Source rank | In / out per M | Context | Listed |
|---|---|---|---|---|---|---|---|---|
| 03 | GPT-6.1 Sol | OpenAI | 1406 | 59.9% | 3 | $2.00 / $10.00 | 1.05M | |
| 22 | GPT-5.5 | OpenAI | 1281 | 56.1% | 23 | $5.00 / $30.00 | 1.05M | |
| 44 | GPT-5.4 | OpenAI | 1228 | 57.6% | 45 | $2.50 / $15.00 | 1.05M | |
| 60 | GPT-5.2 | OpenAI | 1184 | 56.0% | 65 | $1.75 / $14.00 | 400k | |
| 63 | GPT-5 | OpenAI | 1174 | 59.3% | 68 | $1.25 / $10.00 | 400k | |
| 64 | GPT-5.1 | OpenAI | 1168 | 55.9% | 69 | $1.25 / $10.00 | 400k | |
| 72 | GPT-5.3-Codex | OpenAI | 1150 | 51.2% | 78 | $1.75 / $14.00 | 400k | |
| 78 | GPT-5.1-Codex | OpenAI | 1127 | 50.5% | 88 | $1.25 / $10.00 | 400k | |
| 84 | GPT-5 Mini | OpenAI | 1118 | 46.5% | 95 | $0.25 / $2.00 | 400k | |
| 94 | GPT-5.1-Codex-Mini | OpenAI | 1081 | 43.4% | 108 | $0.25 / $2.00 | 400k | |
| 97 | GPT-4.1 | OpenAI | 1068 | 59.1% | 111 | $2.00 / $8.00 | 1.05M | |
| 99 | GPT-4.1 Mini | OpenAI | 1060 | 58.5% | 115 | $0.40 / $1.60 | 1.05M | |
| 102 | GPT-5 Nano | OpenAI | 1034 | 46.5% | 121 | $0.05 / $0.40 | 400k | |
| 104 | o3 | OpenAI | 1023 | 56.9% | 124 | $2.00 / $8.00 | 200k | |
| 107 | o4 Mini (batch) | OpenAI | 992 | 50.0% | 128 | $0.55 / $2.20 | 200k | |
| 108 | gpt-oss-120b | OpenAI | 981 | 40.5% | 129 | $0.15 / $0.60 | 131k | |
| 111 | GPT-4.1 Nano | OpenAI | 959 | 49.6% | 133 | $0.10 / $0.40 | 1.05M | |
| 116 | GPT-4o | OpenAI | 895 | 42.3% | 141 | $2.50 / $10.00 | 128k |
The field and the frontier above are the whole board and do not narrow with this table: a frontier drawn through 18 of 120 models would say nothing beats them on both while something in the same board does. The models this table holds are drawn white in the figures instead.
An Elo from pairwise human preference votes: a higher figure means the model's output was preferred more often against the field it was matched with. Win rate is the share of matchups won. Both shift as new models enter the arena, neither is a measure of correctness, and the source's own rank counts models this catalog does not carry.
Source: OpenRouter (openrouter.ai/models), as of 17 Nov 16:29 UTC. sha256 0bb53a77713f over 853,270 bytes.
Catalog: OpenRouter's public model list.
Intelligence, coding and agentic indexes: Artificial Analysis, as published by OpenRouter.
Elo, win rate and rank: Design Arena, as published by OpenRouter.
Vitna ran none of these evaluations and adjusted none of these figures.
