)
鹈鹕骑自行车大模型测试Pelican Bicycle BenchmarkSVG生成基准测试持续更新评测结果https://github.com/JoJohanse/pelican-bicycle-benchmark测试方法所有模型输入完全相同的提示词generate an svg of a pelican riding a bicycle输出结果按厂商/模型/目录分类存放若无特殊标注均使用模型支持的最大思考强度结果总览共 8 家厂商、14 个模型。厂商按其最早模型的发布时间排列同一厂商内的模型按发布时间先后排序。厂商模型发布时间思考强度输出文件AnthropicClaude Opus 4.72026-04-16maxclaude/claude-opus-4.7/opus-4.7-max.pngAnthropicClaude Opus 4.82026-05-28maxclaude/claude-opus-4.8/opus-4.8-max.jpgAnthropicClaude Fable 52026-06-09maxclaude/claude-fable-5/fable-5-max.jpgAnthropicClaude Sonnet 52026-06-30maxclaude/claude-sonnet-5/sonnet-5-max.pngOpenAIGPT-5.52026-04-23xhighGPT/gpt-5.5/gpt-5.5-pelican-xhigh.pngOpenAIGPT-5.6 Luna2026-07-09high / xhigh / maxGPT/gpt-5.6-luna/OpenAIGPT-5.6 Sol2026-07-09high / xhigh / maxGPT/gpt-5.6-sol/OpenAIGPT-5.6 Terra2026-07-09high / xhigh / maxGPT/gpt-5.6-terra/DeepSeekDeepSeek V4 Flash2026-07-31maxDeepSeek/deepseek-v4-flash/v4-flash.pngDeepSeekDeepSeek V4 Pro2026-08-13maxDeepSeek/deepseek-v4-pro/v4-pro.png小米MiMo v2.5 Pro2026-04-27maxxiaomi/mimo-v2.5-pro/mimo-v2.5-pro.png智谱GLM-5.22026-06-16maxGLM/glm-5.2/glm-5.2-max.jpg智谱GLM-5.32026-08-14maxGLM/glm-5.3/glm-5.3-max.png智谱GLM-5.3 Flash2026-08-26maxGLM/glm-5.3/glm-5.3-flash.png月之暗面Kimi K32026-07-16maxKimi/kimi-k3/kimi-3-pelican.jpg阿里巴巴Qwen 3.8 Max2026-08-03maxQwen/Qwen-3.8-Max/qwen-3.8-max-pelican.png阿里巴巴Qwen 3.8 27B2026-08-14maxQwen/Qwen-3.8-27B/qwen-thinking-bicycle-27b.jpgMetaMuse Spark 1.22026-08-05maxmeta/muse-spark-1.2/muse-spark-1.2.png各模型输出AnthropicClaude Opus 4.72026-04-16Claude Opus 4.82026-05-28Claude Fable 52026-06-09Claude Sonnet 52026-06-30OpenAIGPT-5.52026-04-23xhighGPT-5.62026-07-09DeepSeekDeepSeek V4 Flash2026-07-31DeepSeek V4 Pro2026-08-13小米MiMo v2.5 Pro2026-04-27智谱 GLMGLM-5.22026-06-16GLM-5.32026-08-14GLM-5.3 Flash2026-08-26月之暗面Kimi K32026-07-16阿里巴巴Qwen 3.8 Max2026-08-03Qwen 3.8 27B2026-08-14MetaMuse Spark 1.22026-08-05