分析称 2026 上半年 SoTA 大模型可盈利地以约 $0 输出价服务
有人在 Whale 仓库里挖出线索：下一代 SoTA 模型 100 万上下文输出才 $0.2 还能盈利，信息只发在微信里，看看推理过程。
一位推特用户在 Whale 代码仓库的中文微信群链接中发现线索：2026 上半年的 SoTA 大模型可按 100 万上下文、200+ TPS 的速度提供服务，72 小时缓存保留，缓存命中、输入、输出价格约为 $0、$0.
05、$0.2，且能盈利。
这条信息藏在中文微信渠道里，没有出现在英文公告中。作者认为这体现了中美模型价格认知的信息差。
I mean this stuff. At a glance, just another neat Whale repo. Buried in a link to Zhihu in the Chinese-only WeChat is the payload: H1 2026 SoTA LLM can be served – with 1M context, at 200+TPS, 72 hr cache retention! – for ≈$0|0.05|$0.2 cache/in/out. PROFITABLY. Mutual Knowledge. https://t.co/posEIISXuq
This report is categorized under AITOP's "Models" desk, focusing on China's AI industry and domestic foundation model ecosystem. Within our cross-regional analytical framework, it directly maps to US-China foundation model benchmarks and commercialization comparison and AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include Whale, 推理模型, 开源模型, 定价.
Source attribution: curated from Teortaxes (@teortaxesTex).
