LLM Pricing
← 返回工具

订阅 vs 按量

按你的用量权衡 AI 订阅与按量付费 API。订阅按时间窗口计费而非按 token，因此不存在精确的盈亏平衡点：结果以区间给出，并附缓存敏感度与每个套餐的核对日期。

套餐价格与额度为人工维护，未逐家向官方核实。API 牌价取自目录，因此更换价格基准可能使结论反转。额度规则变动频繁，以官方页面为准。

对话约为 1。Agent 要规划、检索、调工具、执行、总结，一个 prompt 常触发 5–30 次调用。这就是每 token 单价低的模型、每个任务仍可能很贵的原因。

第一方牌价，与第一方订阅属同类可比的替代方案。

这个用量下约 4,800 次模型调用 / 月。

OpenAI Plus最合适额度够用$20.00/月
按此模型计价56.6详情 →$2.00 / $12.00
你的用量按 API 计：$206.64–$245.52/月
订阅每月省 $186.64–$225.52
盈亏平衡约 2 prompt/天缓存 50% 时约 2/天跨宿主价差 3.1×核对于 2026-08-08
这份额度值多少钱 →
单次模型调用18,000 tok · $0.047
官方额度原文~160 messages / 3h
月度容量（最紧窗口：3 小时）75,093–311,467 次调用 · 9,387–38,933 个 prompt
同等用量按官方牌价计$3,536.90–$14,670.08
套餐月费$20.00

这份额度比按 API 买同等用量便宜约 100%。

基于 GPT-5.6 Terra 与左侧画像计算。未计入缓存写入费用，实际 API 成本略高于此。

Anthropic Pro额度够用$20.00/月
按此模型计价55.3详情 →$2.00 / $10.00
你的用量按 API 计：$177.84–$216.72/月
订阅每月省 $157.84–$196.72
盈亏平衡约 2 prompt/天缓存 50% 时约 2/天跨宿主价差 2.1×核对于 2026-08-08
这份额度值多少钱 →
单次模型调用18,000 tok · $0.041
官方额度原文~45 messages / 5h plus a weekly cap
月度容量上限（已知最紧窗口：5 小时）12,672–52,560 次调用 · 1,584–6,570 个 prompt
同等用量按官方牌价计$520.82–$2,160.22
套餐月费$20.00

这份额度比按 API 买同等用量便宜约 99%。

官方口径里还有未公开具体数字的额外上限，因此这里的容量和上面的折扣都是高估值。

基于 Claude Sonnet 5 与左侧画像计算。未计入缓存写入费用，实际 API 成本略高于此。

GitHub Enterprise额度够用$39.00/席/月
按此模型计价63.1详情 →$5.00 / $25.00
你的用量按 API 计：$444.60–$541.80/月
订阅每月省 $405.60–$502.80
盈亏平衡约 1–2 prompt/天缓存 50% 时约 1/天跨宿主价差 2.2×核对于 2026-08-08
这份额度值多少钱 →
单次模型调用18,000 tok · $0.103
官方额度原文3,900 AI credits / user / month
月度容量（最紧窗口：730 小时）31,200–31,200 次调用 · 3,900–3,900 个 prompt
同等用量按官方牌价计$3,205.80–$3,205.80
套餐月费$39.00

这份额度比按 API 买同等用量便宜约 99%。

基于 Claude Opus 5 与左侧画像计算。未计入缓存写入费用，实际 API 成本略高于此。

成本区间来自缓存命中的不确定带（±15 个百分点）；容量区间来自对窗口额度的保守/乐观两种读法。盈亏平衡是一个区域而不是一个点，因为订阅额度按时间窗口计，折不成固定的 token 数。

短于一天的滚动窗按每月 176 个活跃小时（8 小时 × 22 天）折算；日/周/月额度属日历窗口，不打折。

无法计算额度的套餐

这些套餐没有公布绝对额度，或计价单位无法折算成模型调用次数。这里如实列出，而不是给一个编造的数字。

  • OpenAI Free免费Free-tier limits vary by model and are not published as absolute numbers.
  • OpenAI Go$8.00Higher than Free but the allowance is not published as an absolute number.
  • OpenAI Business$25.00Per-seat allowance follows the Pro tier but is not published as an absolute number.
  • Anthropic Free免费Free-tier session limits are not published as absolute numbers.
  • Anthropic Team$25.00Per-seat allowance is not published as an absolute number.
  • Google AI Plus$4.99Entry-tier allowance is not published as an absolute number.
  • Google AI Pro$19.99Published only as ~4× the free tier, and the free tier has no absolute number — so no month cap can be derived.
  • Google AI Ultra$249.99Highest-tier allowance is not published as an absolute number.
  • xAI SuperGrok Lite$10.00Rate limits are not published as absolute numbers.
  • xAI SuperGrok$30.00Rate limits are not published as absolute numbers.
  • xAI SuperGrok Heavy$300.00Rate limits are not published as absolute numbers.
  • Perplexity Free免费A few Pro searches per day; the standard-search allowance is not published.
  • Perplexity Pro$20.00Quota is in Pro searches, which bundle retrieval plus generation and can't be mapped to model calls.
  • Perplexity Max$200.00Quota is in Pro searches, which bundle retrieval plus generation and can't be mapped to model calls.
  • Mistral Free免费Basic limits are not published as absolute numbers.
  • Mistral Pro$14.99Higher limits are not published as absolute numbers.
  • Mistral Team$24.99Per-seat allowance is not published as an absolute number.
  • GitHub Free免费Monthly completion and chat caps are tier-specific and not published as premium-request counts.
  • GitHub Pro$10.00Official pages state a premium-request quota without an absolute monthly number.
  • GitHub Pro+$39.00Official pages state the highest premium-request quota without an absolute monthly number.
  • GitHub Business$19.00Per-seat premium-request quota is not published as an absolute number.
  • Cursor Hobby免费Agent-request limits are not published as absolute numbers.
  • Cursor Pro$20.00Bills as a $20 credit of included API usage rather than a request or token cap — compare with the cost calculator instead.
  • Cursor Pro+$60.00Bills as included API usage rather than a request or token cap.
  • Cursor Ultra$200.00Bills as included API usage rather than a request or token cap.
  • Cursor Teams$40.00Pooled per-seat usage rather than a request or token cap.