Output Loop Bench · Artifacts

Logo: AI models compared

A logo in SVG, shown in its main versions. 30 models tested on the same task.

Landscaper logo (3 versions)

Logo design from code is a hard, visible test of taste and restraint. A good mark must still work small and in one colour.

ModelMakerTimeTokens
claude-haiku-5-5 ★ Output Loop's pickAnthropic67.3s100,363
composer-2.5Cursor21.6s23,765
muse-spark-1.3-mediumMeta28.8s26,594
nemotron-3-nano-4bNVIDIA38.6s4,106
MiniMax-M2.5MiniMax40.4s18,900
gemini-4-argon-mediumGoogle57.7s22,653
claude-fable-5-1Anthropic65.5s96,521
gpt-6-astraOpenAI66.2s17,437
glm-5.3Z.ai67.3s21,147
gpt-6.1-solOpenAI73.0s17,891
gemma-4-26b-a4b-qatGoogle74.0s8,907
claude-sonnet-5-5Anthropic92.6s101,308
kimi-k3-maxMoonshot (Kimi)93.1s36,782
bonsai-27bPrism ML108.0s6,096
claude-opus-5-5Anthropic120.6s101,195
gpt-6-lunaOpenAI125.7s28,927
gemini-3.1-proGoogle135.1s22,278
deepseek-v4-proDeepSeek142.7s33,864
kimi-k2.7-codeMoonshot (Kimi)146.5s42,095
qwen3.7-plusAlibaba (Qwen)152.3s25,327
muse-glimmerMeta161.4s5,310
qwen3.8-flashAlibaba (Qwen)223.3s29,384
muse-glimmer-30bMeta225.2s20,443
kimi-k3Moonshot (Kimi)246.7s18,994
nemotron-3.5-lightning-30b-a3bNVIDIA253.4s22,356
deepseek-v4.1-flashDeepSeek285.1s44,621
gemini-3.8-flashGoogle289.7s53,262
grok-4.7-build-fastxAI340.7s51,776
grok-4.7xAI356.0s43,556
qwen3.8-maxAlibaba (Qwen)563.5s38,968

Open Logo in the interactive Bench