创建

登录 ReadmeX

或

了解我们如何处理你的信息: 隐私政策

ReadmeX

陪你读懂 AI 动态

今天,想先了解什么?

选一份动态总结,或直接聊聊你的问题。

社区和关注动态总结只属于你自己。

资讯

OpenRouter测试:fp8、fp4与fp16端点未检出准确率差异

AI 总结

OpenRouter于2026年10月20日分析了1,455个提供商端点,发现其中865个(59.5%)未报告精度标签。在当天对六个Qwen3.8 27B端点的测试中,240项任务未显示fp8、fp4与fp16之间存在准确率差异;六个端点的得分均在98.3%至99.2%之间。另有40项乘法任务中,一个fp4端点得分最低,但差距小到无法排除偶然因素。较低精度通常价格更低,但并非总是如此:8-bit在24个可比模型中有14个比16-bit便宜,4-bit在33个模型中有26个比8-bit便宜。质量比较仅覆盖一个模型和一天,OpenRouter也无法确定各提供商采用的量化方法。

深度见解

在这一有限测试中,精度标签未显示出对准确率的预测能力,而即使精度相同,不同端点的价格差异也很大。OpenRouter建议用户在采用量化选项前,先使用自己的提示词和工作负载测试候选端点。

LLM Quantization: What fp8 and fp4 Do to Quality and Price

OpenRouter ·11/11/2026

LLM Quantization: What fp8 and fp4 Do to Quality and Price
LLM Quantization: What fp8 and fp4 Do to Quality and Price

When you call an open-weight model through an API, several providers may serve it at different precisions, such as 16-bit bf16, 8-bit fp8 or 4-bit fp4. The precision can affect both the answers and the price. LLM quantization creates these versions by storing a model’s weights in fewer bits, so the model needs less memory and hardware, at some cost in accuracy.

This post explains the formats and reports what we measured on OpenRouter on October 20, 2026, across 1,455 provider endpoints and six endpoints serving Qwen3.8 27B. On our test, we didn’t detect an accuracy difference between fp8, fp4 and fp16, so the precision label alone didn’t predict accuracy. Lower precision was usually cheaper, but not for every model.

#OpenRouter#Qwen3.8 27B#gpt-oss-120b#MiniMax M3#Kimi K2.6#Gemma 4 31B

原文约 12 分钟读完阅读原文
群聊

还没有评论,来说说你的看法。