Netlify与OpenRouter建立合作伙伴关系，推出AI Gateway和Agent Runners两项新功能，以扩展用户对多款AI模型的选择[1]。这两项功能支持包括Claude、GPT、Gemini、Kimi、GLM、DeepSeek等在内的多个模型[1]。
为了评估不同模型的实际表现，Netlify对11个AI模型运行同一提示进行测试对比，衡量它们在生成咖啡店网站时的性能、设计质量和信用消耗[1]。测试涵盖多个场景，包括邻近咖啡店单页网站、待办事项列表应用以及"我能做什么"食谱生成应用[1]。结果显示模型在功能、成本和质量上存在显著差异[1]。
在信用消耗方面，差距尤其明显[1]。Claude Opus 5在一次运行中消耗1,055信用，约为其他模型的4倍，而其他模型平均消耗从2.4信用（DeepSeek V4 Flash 0731）到141信用（GPT 5.6 Sol）不等[1]。根据Netlify的定价体系，免费计划用户拥有300信用，个人计划包含1,000信用，专业计划包含3,000信用，额外信用包为$10/1,500信用[1]。Netlify采用自主开发并开源的AXIS工具进行模型评估，重点关注生成网站的功能正确性而非设计[1]。
Netlify has partnered with OpenRouter to expand AI model selection capabilities, introducing two new features: AI Gateway support for any model on OpenRouter, and Agent Runners that extend access to cutting-edge coding models including Kimi K3, GLM 5.2, and DeepSeek V4.[1] To demonstrate the practical implications of this expanded choice, Netlify conducted a comparative test running identical prompts across 11 AI models to generate a website for a local coffee shop, revealing significant disparities in performance, design quality, and resource consumption.[1]
The cost differences among models proved stark during testing.[1] Claude Opus 5 consumed 1,055 credits in a single run—approximately four times more than competing models—while other models ranged from 2.4 credits for DeepSeek V4 Flash 0731 to 141 credits for GPT 5.6 Sol, with most averaging between 143 and 253 credits.[1] Netlify's credit structure reflects these variations: free plan users receive 300 credits, personal plans include 1,000 credits, professional plans provide 3,000 credits, and additional credit packages cost $10 per 1,500 credits.[1] The company employed its own open-sourced AXIS evaluation tool to assess model performance, focusing on the functional correctness of generated websites rather than design aesthetics.[1]