问题
多个信号反复描述同一类人：要在 Anthropic 和 OpenAI 之间做选型决策的 "developers and teams building on frontier LLM APIs"。9 月下旬一周内模型经济性剧烈变化——Claude Opus 5.5 比 Opus 5 便宜 40%、GPT-6 Sol/Luna 比 5.6 版本便宜 50%、GPT-6.1-Sol 价格仅为 Astra 的五分之一、缓存读取便宜 60%——选型信息随时过期，靠人工追新闻和手动比价越来越难。
目标用户
在 LLM API 上构建应用的开发者和 AI 产品团队，尤其是需要控制推理成本的中大型项目；信号多次出现 "AI product builders and developers choosing frontier models"，属于规模庞大但付费意愿尚不明确的开发者群体。
解决方案
- 价格与跑分聚合：收录 Claude、GPT 等系列的每百万 token 输入/输出价格与基准成绩，自动随发布更新（如 GPT-6 Luna 的 $0.10/$0.50 per M）。
- 成本计算器：输入预估 token 用量和任务类型，输出各模型的月成本对比，含缓存读取折扣。
- 降价/新模型提醒：模型价格或能力发生显著变化时推送，避免团队继续用过期选型。
- AI 选型推荐：用 LLM 根据任务描述（如长程 agent 对话、日常 bug 修复）推荐性价比最高的模型并给出理由。
- 选型决策快照：记录团队当前用的模型与当时的价格，方便事后复盘换模型能省多少。
为什么是现在
2026 年 9 月下旬一周内密集出现大规模降价与换血：Opus 5.5 便宜 40%、GPT-6 Sol/Luna 便宜 50%、GPT-6.1-Sol 仅为 Astra 价格的五分之一、缓存读取便宜 60%，Simon Willison 称 GPT-6 Luna "half the price of that again"。模型经济性的变化速度首次超过了团队人工跟进的速度，选型决策从一次性变成了需要持续监控的事情。
MVP 范围
第一版做：聚合主流厂商的模型价格、跑分和发布动态；按用户的任务类型和月 token 用量算出各方案成本并排序；关键模型降价或新模型上线时推送提醒。不做：实际代理流量或统一 API 网关、自动切换生产环境模型、私有部署评测。
风险
- 平台依赖：数据完全来自模型厂商的发布与定价，厂商改版或封锁抓取会直接断粮。
- 时效风险：价格一周内可降 40%-50%，数据稍有延迟就失去核心价值，需要高度自动化的更新管道。
- 上游吞并：模型聚合平台或云厂商（如已有 Claude Platform API 这类入口）随时可能内置比价和路由功能。
- 付费意愿未验证：信号里只有模型价格，没有用户为选型工具掏钱的证据，可能最终只是流量型内容站。
信号证据
这张卡片依据的原始讨论。摘录保持原文，点“原文”查看上下文。
Simon Willison's Weblog9月29日新品发布
“GPT 6.1 Sol: Near-Astra intelligence for a fifth of the price”
OpenAI launched GPT-6.1-Sol, offering near-Astra-level intelligence at about a fifth of the price of comparable models.原文
Simon Willison's Weblog9月28日新能力隐含付费意愿
“They say it "runs 30%+ faster, and costs up to 30% less for most work" - it's priced the same as Sonnet 5 but appears to beat it on every benchmark, and should be cheaper to run as well.”
Anthropic released Claude Sonnet 5.5, a new model that runs 30%+ faster, costs up to 30% less, and beats Sonnet 5 on every benchmark.原文
The Rundown AI9月23日新品发布隐含付费意愿
“GPT-6 Sol and Luna come in with slight score increases over their 5.6 counterparts, but are 50% cheaper at $0.10/$0.50 per M (Luna) and $2/$10 (Sol).”
AI model buyers can now get frontier-level intelligence at significantly reduced prices, with Anthropic releasing Claude Opus 5.5 at 40% less price than its predecessor and OpenAI releasing GPT-6 Sol and Luna at half the price of the versions they replace.原文
Latent Space9月23日新能力隐含付费意愿
“OpenAI made a valiant effort with GPT-6 Sol and Luna launching 50% lower than GPT-5.6, but with 17M views on the launch and counting, today was always going to belong to Claude Opus 5.5”
OpenAI launched GPT-6 Sol and Luna priced 50% lower than GPT-5.6, continuing aggressive frontier model price cutting.原文
Latent Space9月23日新品发布隐含付费意愿
“Opus 5.5 “performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5” (@claudeai; @AnthropicAI).”
Anthropic shipped Claude Opus 5.5, a flagship model delivering near-frontier capability at 40% lower cost, with improved writing and instruction following.原文
Simon Willison's Weblog9月22日新能力隐含付费意愿
“GPT-5.6 Luna was already my favorite model for building applications against, because it combined excellent performance with being really cheap. Somehow GPT-6 Luna is half the price of that again”
OpenAI's GPT-6 Luna halves pricing versus GPT-5.6 Luna to $0.10/M input and $0.50/M output, making cheap application-building models dramatically cheaper.原文
Simon Willison's Weblog9月22日新能力隐含付费意愿
“Opus 4.5, 4.6, 4.7, 4.8, and 5 all shared the same price: $5/million tokens for input and $25/million for output. 5.5 is a 20% reduction - $4/million and $20/million.”
Anthropic released Claude Opus 5.5 with a 20% price cut and 60% cheaper cache reads, aimed at longer agentic conversations.原文