Anthropic 发布 Claude Haiku 5.5——其最便宜、最快的小模型，按官方算法平均比 Haiku 4.5 便宜约 75%，10 万 token 以内每百万 token 输入 $0.10／输出 $0.50——同时 Sonnet 5.5 缓存读取价格砍半，并为 Max 与 Team 宣布每月 API 额度

“Anthropic 新发的 Haiku 5.5 按官方算法平均比 Haiku 4.5 便宜约 75%——官方说它就是那个当便宜子智能体的料：重复的活它来干，Opus 和 Sonnet 负责规划。”

可用说法

• 2026 年 10 月 7 日（星期三，已按日历核算），Anthropic 发布 Claude Haiku 5.5：「the cheapest, fastest, and most capable small model we've ever released（我们发布过的最便宜、最快、最强的小模型）。」按公告：「Claude Haiku 5.5 is designed for high-volume, cost-sensitive tasks. It reliably handles quick and repetitive workloads (like summaries, compactions, database queries, and classification requests). It pairs well with Opus 5.5 and Sonnet 5.5 as a subagent on coding work. And, since it's also our fastest model to date, it works especially well for speed-sensitive tasks like live customer support and browser use.（Claude Haiku 5.5 为高流量、成本敏感的任务设计。它稳定处理快速、重复的工作负载（如摘要、压缩、数据库查询和分类请求）。它与 Opus 5.5 和 Sonnet 5.5 在编码工作上搭配得当，可作为子智能体。而且，因为它也是我们迄今最快的模型，它特别适合速度敏感任务，如实时客服与浏览器使用。）」最快模型的宣称带着 Anthropic 自己的脚注：「Claude Haiku 5.5 is our fastest model to date at each model's standard speed, although it runs less quickly than our Opus models in Fast Mode.（Claude Haiku 5.5 是我们迄今最快的模型——以各模型的标准速度计；在 Fast Mode 下它跑得比我们的 Opus 模型慢。）」定价，按公告：「Haiku 5.5 is available at a much lower price than Haiku 4.5. On average, it now costs around 75% less to run.（Haiku 5.5 的价格远低于 Haiku 4.5。平均而言，现在跑起来便宜约 75%。）」脚注给这个平均数划了范围：「Claude Haiku 5.5 is priced 90% lower than Claude Haiku 4.5 for requests up to 100,000 tokens, and 50% lower for requests over 100,000 tokens. On Haiku 4.5, 90% of requests fell into the former category. This calculation also accounts for changes between Haiku 4.5 and Haiku 5.5 in how many tokens are used to complete a given piece of work: Haiku 5.5 has an updated tokenizer (similar to Sonnet 5.5's and Opus 5.5's), which means it uses slightly more tokens per task.（10 万 token 以内的请求，Claude Haiku 5.5 比 Claude Haiku 4.5 定价低 90%；超过 10 万 token 的请求低 50%。在 Haiku 4.5 上，90% 的请求属于前一类。该计算也把 Haiku 4.5 与 Haiku 5.5 在完成同一件工作所用 token 数量上的变化算了进去：Haiku 5.5 用了新的分词器（与 Sonnet 5.5、Opus 5.5 的类似），意味着每件任务用的 token 略多。）」按定价表，Haiku 5.5 在 prompt 不超过 10 万 token 时为每百万 token 输入 $0.10／输出 $0.50（超过 10 万为 $0.50／$2.50），缓存读取 $0.01／$0.05。按公告：「Haiku 5.5 is especially good value when used for tasks with prompts up to 100,000 tokens, which make up around 90% of requests to our previous Haiku model.（对 prompt 不超过 10 万 token 的任务，Haiku 5.5 特别划算——这类请求约占上一代 Haiku 模型请求量的 90%。）」Haiku 5.5 是「our first Haiku-class model to come with an adjustable effort setting（我们第一个带可调 effort 设置的 Haiku 级模型）」，按公告，用户「can decide whether to optimize for cost or intelligence（可以决定为成本还是智能优化）」。Anthropic 报告的基准：OSWorld 2.1（离线子集）72.4%，对 Haiku 4.5 的 15.7%；Humanity's Last Exam 无工具 45.9%，对 Haiku 4.5 的 10.2%；Terminal-Bench 4.0 39.2%，对 Sonnet 5.5 的 70.6%——公告自己也给模型定了位：「Sonnet 5.5 and Opus 5.5 remain better choices for complex agentic coding tasks like those measured by Terminal-Bench 4.0. By contrast, Haiku 5.5 is best suited to more narrowly scoped tasks that might otherwise have been cost-prohibitive with previous versions of Claude—like compaction, summarization, or subagent work.（对 Terminal-Bench 4.0 所测的那种复杂 agentic 编码任务，Sonnet 5.5 与 Opus 5.5 仍是更好的选择。相比之下，Haiku 5.5 最适合范围更窄的任务——压缩、摘要或子智能体工作——它们在以前版本的 Claude 上可能贵到不划算。）」安全方面，按公告，Haiku 5.5 的网络安全防护「more restrictive than Haiku 4.5's, but somewhat less restrictive than those we've applied to other recent models（比 Haiku 4.5 的更严格，但比我们近来对其他模型施加的略宽松）」，允许比 Sonnet 5.5 防护「a wider range of defensive tasks（更广的防御性任务）」，但仍「still block penetration testing and other techniques more likely to be used by attackers（仍会拦截渗透测试及其他更可能被攻击者利用的技术）」。可用性，按公告：「Claude Haiku 5.5 is available now on all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure.（Claude Haiku 5.5 现已在所有平台可用，包括 Amazon Web Services、Google Cloud 与 Microsoft Azure。）」且按公告：「For developers, we're also updating our Claude Python and TypeScript SDKs to add support for computer use and browser use in beta.（对开发者，我们还在更新 Claude Python 与 TypeScript SDK，以 beta 形式加入对计算机使用与浏览器使用的支持。）」随模型一起，Anthropic 宣布：「First, starting today, we're lowering the price of cache reads on Claude Sonnet 5.5. Cache reads now cost 50% less: $0.10 per million tokens rather than $0.20.（首先，从今天起，我们下调 Claude Sonnet 5.5 的缓存读取价格。缓存读取现在便宜 50%：每百万 token $0.10，而不是 $0.20。）」——Anthropic 称这让「Sonnet 5.5 on most agentic tasks」的成本「by around 20%（降低约 20%）」。以及：「Second, this week, we'll roll out a new monthly API credit to all Max and Team subscribers for use on the Claude Platform. Max 5x users will get $100 in credits per month, Max 20x users will get $200, and Team subscribers will receive up to $500, pooled across their users.（其次，本周我们将向所有 Max 与 Team 订阅者推出新的每月 API 额度，用于 Claude Platform。Max 5x 用户每月 $100 额度，Max 20x 用户 $200，Team 订阅者最高 $500、在其用户间共享。）」——额度可用于「any of our models（我们的任何模型）」。按 Simon Willison（10 月 7 日发文）：「The new Haiku exactly matches the price of GPT-6 Luna—$0.10/$0.50—up to 100,000 tokens.（新 Haiku 在 10 万 token 以内与 GPT-6 Luna 完全同价——$0.10/$0.50。）」Willison 还测量到：「My Claude Token Counter tool shows that the same long prompt uses around 1.25x as many tokens with Haiku 5.5 compared to Haiku 4.5, so there's a hidden price increase there.（我的 Claude Token Counter 工具显示，同一个长 prompt 在 Haiku 5.5 上比 Haiku 4.5 所用 token 约为 Haiku 4.5 的 1.25 倍，所以那里藏着一层变相涨价。）」关于额度，按 Willison：「The API credits exactly match the cost of the subscription itself. This is really generous—it makes it much easier for subscribers to use the API.（这些 API 额度恰好等于订阅本身的价格。这真的很大方——它让订阅者用 API 容易多了。）」以及：「Note that the monthly credits do not roll over—use them or lose them.（注意，每月额度不滚存——用不掉就没了。）」Anthropic 页面转述的客户报告结果：Asana（Aaron Vinh）看到「over a 30% reduction in latency for task completions and up to 2.5x faster inference per agent turn（任务完成延迟降低超 30%，每个 agent 轮次推理最快快 2.5 倍）」；HubSpot（Ze'ev Klapow）报告其 CRM 套件上「92.8% averaged over three runs（三次运行平均 92.8%）」；AlphaSense（Daniel Campos）的「Ask in Document」在生产环境「about 8M calls a week in production（每周约 800 万次调用）」，对比 Haiku 4.5 测得「0.84 vs. 0.76」；Cognition（Walden Yan）报告在 Devin Fusion 中以 Haiku 5.5 作 sidekick 拿到「a top-tier FrontierCode score of 66.2（66.2 的顶级 FrontierCode 分数）」。