← All Tools ← 全部工具 🎮 小游戏
Home首页 › AI Tools AI 工具 › Claude Haiku 5.5
🧠 LLM API 大模型 API 🌟 Anthropic · 2026-10-07 发布 · 史上最便宜最快的小模型（2026-10 核实） anthropic llm-api claude fast-model claude-haiku-5-5 subagent

Claude Haiku 5.5 – $0.10 超低价极速小模型：1M 上下文、可调 effort，高并发 Agent 最佳副驾

Claude Haiku 5.5 is Anthropic's small model in the Claude 5.5 family, released October 7, 2026 — "the cheapest, fastest, most capable small model we've ever released": 1M context, 128K max output, first Haiku with adjustable effort. API pricing is $0.10/$0.50 per MTok for prompts up to 100K tokens ($0.50/$2.50 above), ~75% cheaper than Haiku 4.5 on average; prices per official siteClaude Haiku 5.5 是 Anthropic 2026-10-07 发布的 Claude 5.5 家族小模型——官方称其为「史上最便宜、最快、能力最强的小模型」：1M 上下文、12.8 万最大输出、Haiku 系列首个可调 effort。API 价格 $0.10/$0.50 每百万 token（prompt ≤100K；超 100K 按 $0.50/$2.50 计费），比 Haiku 4.5 平均便宜约 75%；价格以官网为准

🗓 Updated更新:
Visit Official Site ↗ 访问官网 ↗ ⚖️ Compare
Category分类
LLM API大模型 API
llm-api
Open Source开源状态
闭源（API/云平台）
Claude API, AWS, Google Cloud, Azure · model ID claude-haiku-5-5Claude API、AWS、Google Cloud、Azure · 模型 ID claude-haiku-5-5
License许可证
Proprietary（专有）
API $0.10/$0.50 per MTok (≤100K); $0.50/$2.50 above — prices per official siteAPI $0.10/$0.50 每百万 token（≤100K）；超 100K 按 $0.50/$2.50；价格以官网为准
Tags标签
anthropic, llm-api, claude, fast-model, claude-haiku-5-5, subagent
6 tags total个标签

What Is Claude Haiku 5.5?Claude Haiku 5.5 是什么？

Claude Haiku 5.5 is Anthropic's small, fast, cheap model in the Claude 5.5 family, released on October 7, 2026. In Anthropic's own words, it is "the cheapest, fastest, and most capable small model we've ever released" — built for high-volume, cost-sensitive work like summaries, compactions, database queries and classification, and positioned as the ideal subagent running under Opus 5.5 or Sonnet 5.5 (the lead model plans, Haiku 5.5 executes the speed-sensitive chores). Specs: 1M-token context (up from 200K), 128K max output, a June 2026 knowledge cutoff, and — for the first time in the Haiku line — adjustable effort (low, medium, high, xhigh, max; default medium).Claude Haiku 5.5 是 Anthropic 2026 年 10 月 7 日发布的 Claude 5.5 家族小模型。官方原话：「史上最便宜、最快、能力最强的小模型」——为高频、低成本任务而生：摘要、compaction、数据库查询、分类，也被定位为 Opus 5.5 / Sonnet 5.5 之下的理想 subagent（大模型做规划，Haiku 5.5 干速度敏感的脏活累活）。参数：100 万上下文（从 20 万升级）、12.8 万最大输出、2026 年 6 月知识截止，Haiku 系列首次支持可调 effort（low/medium/high/xhigh/max，默认 medium）。

Pricing is $0.10 input / $0.50 output per million tokens for prompts up to 100,000 tokens. Watch the cliff: cross 100K tokens and the whole request reprices fivefold — $0.50 input / $2.50 output per MTok. Anthropic says it is 90% cheaper than Haiku 4.5 below the threshold, 50% cheaper above it, about 75% cheaper on a real workload. Prices are per the official site.定价 $0.10 输入 / $0.50 输出每百万 token，但只适用于 100K 以内的 prompt。注意价格悬崖：超过 100K，整单 5 倍跳涨——$0.50 输入 / $2.50 输出每百万 token。Anthropic 称：100K 内比 Haiku 4.5 便宜 90%，100K 外便宜 50%，真实负载平均便宜约 75%。价格以官网为准。

⚠️ Price-cliff warning: the $0.10/$0.50 headline rate applies only to prompts of 100,000 tokens or fewer. Long agent histories and repo-wide context cross that line easily — keep prompts compact or pay the second tier ($0.50/$2.50).价格悬崖提醒：$0.10/$0.50 的 headline 价只适用于 100K 以内的 prompt。长 Agent 历史和仓库级上下文很容易越线——要么压紧 prompt，要么接受第二档（$0.50/$2.50）。

Haiku 5.5 is Anthropic finally playing the price war properly: $0.10/$0.50 under 100K tokens is a direct match to OpenAI's GPT-6 Luna, and Anthropic's own table has Haiku 5.5 beating Luna on every shared benchmark — most dramatically 72.4% vs 48.9% on OSWorld desktop tasks. The adjustable effort setting is the genuinely new thing: a per-request dial for cheap-vs-smart that makes Haiku 5.5 a real routing layer, not just a dumb cheap model. The honest caveats: the 100K price cliff is brutal and easy to cross by accident, the benchmark numbers are all Anthropic-reported and unverified, and Haiku 4.5's old 200K context limit doubling to 1M doesn't help you if your prompt lands in the expensive tier. Verdict: the best-value Anthropic model for high-volume classification/extraction/subagent work today — but wire in prompt-length guards before you route real traffic.

Haiku 5.5 是 Anthropic 第一次真正打价格战：$0.10/$0.50（100K 内）直接对标 OpenAI GPT-6 Luna，而且官方表格里 Haiku 5.5 在所有共享基准上都赢了 Luna——OSWorld 桌面任务 72.4% 对 48.9%，最悬殊。可调 effort 是真正的创新：每次调用都能拨「便宜还是聪明」的旋钮，让 Haiku 5.5 成了真正的路由层，不只是个傻便宜的模型。真实代价：100K 价格悬崖很陡、很容易无意越线；基准数据全是 Anthropic 自报、未经验证；上下文从 20 万翻到 100 万很香，但 prompt 落进贵档就白搭。结论：今天 Anthropic 家做高频分类/抽取/subagent 性价比最高的选择——但接真实流量前先把 prompt 长度守卫接上。

— 中国吧 AI Tools Hub Editorial Team

Who Should Use Claude Haiku 5.5?谁适合使用Claude Haiku 5.5？

✓ Good Fit For适合以下场景

  • High-volume classification, extraction, summarization at $0.10/$0.50高频分类、抽取、摘要——$0.10/$0.50 的价格
  • Subagents under Opus/Sonnet 5.5 for speed-sensitive choresOpus/Sonnet 5.5 之下的 subagent，干速度敏感的活
  • Live support and browser-use tasks (OSWorld 72.4%)在线客服、浏览器操作类任务（OSWorld 72.4%）

✕ Not Ideal For不适合以下场景

  • Open-ended coding agents — keep Sonnet 5.5 as the default (70.6% vs 39.2% on Terminal-Bench)开放式编程 Agent——Sonnet 5.5 仍是默认（Terminal-Bench 70.6% 对 39.2%）
  • Long-context workloads that cross the 100K-token price cliff会越过 100K 价格悬崖的长上下文负载

Getting Started with Claude Haiku 5.5Claude Haiku 5.5 快速开始 

1. Use model ID claude-haiku-5-5 on the Claude API, Amazon Bedrock, Google Cloud or Microsoft Foundry — also in the Claude.ai model picker (Free plan included) and Claude Code via /model claude-haiku-5-5在 Claude API、Amazon Bedrock、Google Cloud、Microsoft Foundry 用模型 ID claude-haiku-5-5——Claude.ai 模型选择器（含免费档）和 Claude Code（/model claude-haiku-5-5）也有
2. Budget $0.10 input / $0.50 output per MTok for prompts ≤100K; $0.50/$2.50 above — prices per official siteprompt ≤100K 按 $0.10 输入 / $0.50 输出每百万 token 做预算；超 100K 按 $0.50/$2.50——价格以官网为准
3. Set effort per request (low/medium/high/xhigh/max, default medium); for subagent fleets pair Haiku 5.5 with Opus/Sonnet 5.5 as the lead model每次调用按需设 effort（low/medium/high/xhigh/max，默认 medium）；做 subagent 舰队时，用 Opus/Sonnet 5.5 当 lead 模型、Haiku 5.5 当执行层
💡 Tip: add a prompt-length guard to your router — one 100K+ prompt silently quintuples the bill. Compare the top two effort levels against Sonnet 5.5 (Anthropic's own advice) before using max effort on hard tasks.小技巧：给路由加上 prompt 长度守卫——一条 100K+ 的 prompt 会无声无息地让账单翻 5 倍。难任务用 max effort 之前，先按官方建议拿它和 Sonnet 5.5 对比测一下。

Key Features核心功能

  • 💰
    $0.10/$0.50 Headline Pricing$0.10/$0.50 超低价 — $0.10 input / $0.50 output per million tokens for prompts up to 100K tokens — 90% cheaper than Haiku 4.5 below the line, ~75% cheaper on real workloads. Prices per official site.$0.10 输入 / $0.50 输出每百万 token（prompt ≤100K）：100K 内比 Haiku 4.5 便宜 90%，真实负载平均便宜约 75%。价格以官网为准。
  • 🎛️
    Adjustable Effort (First for Haiku)可调 effort（Haiku 系列首次） — Per-request dial: low, medium (default), high, xhigh, max — trade latency and cost against quality on every call.每次调用可拨：low/medium（默认）/high/xhigh/max，在延迟、成本与质量之间取舍。
  • 📏
    1M Context, 128K Output100 万上下文、12.8 万输出 — 1M-token context window (up from 200K) with up to 128K output tokens — but mind the 100K price cliff.100 万上下文（从 20 万升级）、输出最高 12.8 万——但注意 100K 价格悬崖。
  • 🖥️
    Strong Computer/Terminal Use强桌面/终端操作能力 — Anthropic-reported 72.4% on OSWorld 2.1 offline and 39.2% on Terminal-Bench 4.0 (Haiku 4.5 scored 0.0% there) — a leap for the small tier.官方报告 OSWorld 2.1 离线 72.4%、Terminal-Bench 4.0 39.2%（Haiku 4.5 那项是 0.0%）——小模型档的一次飞跃。
  • 🌐
    Broad Availability多渠道可用 — Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, Claude Code, and every Claude plan including Free. See also Claude and Claude Sonnet 5.5.Claude API、Amazon Bedrock、Google Cloud、Microsoft Foundry、Claude Code，以及 Claude 各档计划（含免费版）。另见 Claude 与 Claude Sonnet 5.5。

Pros & Cons优缺点

✓ Good Fit For适合以下场景

  • Cheapest Anthropic model ever — $0.10/$0.50 matches GPT-6 Luna's headline rateAnthropic 史上最便宜——$0.10/$0.50 直接对标 GPT-6 Luna
  • First Haiku with adjustable effort — a real per-request cheap-vs-smart dialHaiku 系列首次可调 effort——真正的「便宜还是聪明」旋钮
  • Huge capability jump over Haiku 4.5 (OSWorld 72.4% vs 15.7%)比 Haiku 4.5 能力飞跃（OSWorld 72.4% 对 15.7%）
  • On the Free plan and Claude Code — cheap to try免费档和 Claude Code 都能用，试用成本极低

✕ Not Ideal For不适合以下场景

  • The 100K-token price cliff (5x repricing) is easy to cross by accident100K 价格悬崖（5 倍跳涨）很容易无意越过
  • All benchmark numbers are Anthropic-reported, unverified基准数据均为 Anthropic 自报，未经验证
  • Still lags the big models on hard multi-step coding (39.2% vs 70.6%)硬核多步编程仍落后大模型（39.2% 对 70.6%）

Use Cases应用场景

Popular ways people use Claude Haiku 5.5.人们使用Claude Haiku 5.5的常见场景：

🤖 Subagent FleetsSubagent 舰队

Opus/Sonnet 5.5 plans and reasons, Haiku 5.5 runs the speed-sensitive chores underneath — the two-model workflow Anthropic is pushing across its lineup.Opus/Sonnet 5.5 做规划和推理，Haiku 5.5 在下层干速度敏感的活——Anthropic 正在全系推的双模型工作流。

🏷️ Classification & Extraction分类与抽取

High-volume labeling, moderation and data extraction at $0.10/$0.50 per MTok — the core "cheapest model" workload.高频打标、审核、数据抽取，$0.10/$0.50 每百万 token——「最便宜模型」的核心场景。

💬 Live Support & Browser Tasks在线客服与浏览器任务

72.4% on OSWorld 2.1 offline makes it the strongest Haiku ever for computer-use and live customer support at this price.OSWorld 2.1 离线 72.4%，是这个价位做计算机操作和在线客服最强的 Haiku。

📝 Summarization at Scale规模化摘要

Summaries and compactions by the million — set effort per request and keep prompts under 100K to stay in the cheap tier.百万级摘要与 compaction——每次调用按需设 effort，prompt 压在 100K 以内留在廉价档。

Similar AI Tools相似 AI 工具

If Claude Haiku 5.5 doesn't fit your needs, here are other popular AI tools you might consider — including the rest of the family: Claude and Claude Sonnet 5.5 (see our full reviews).如果Claude Haiku 5.5不符合你的需求，可以看看这些热门 AI 工具——包括同家族的 Claude 和 Claude Sonnet 5.5（见本站完整评测）。

Claude
Anthropic's flagship assistant — conversation, docs, slides, designAnthropic 旗舰对话助手：对话、文档、幻灯片、设计
Claude Sonnet 5.5
The balanced 5.5 workhorse — $2/$10, cyber safeguards, distillation defense5.5 家族均衡主力：$2/$10、网络安全防护、防蒸馏
Claude Code
Anthropic's terminal agent — Opus 5.5 default, 1M contextAnthropic 终端 Agent：Opus 5.5 默认、100 万上下文
Cursor
AI-native IDE with agent workflowsAI 原生 IDE，带 Agent 工作流

Frequently Asked Questions常见问题

What is Claude Haiku 5.5?Claude Haiku 5.5 是什么？ ▼
Claude Haiku 5.5 is Anthropic's small, fast, cheap model in the Claude 5.5 family, released on October 7, 2026. Anthropic calls it the cheapest, fastest, most capable small model it has ever released: 1M-token context, 128K max output, a June 2026 knowledge cutoff, and the first Haiku-class model with an adjustable effort setting. It is designed as a high-volume workhorse and as a subagent under Opus 5.5 and Sonnet 5.5.Claude Haiku 5.5 是 Anthropic 2026 年 10 月 7 日发布的 Claude 5.5 家族小模型。官方称其为「史上最便宜、最快、能力最强的小模型」：100 万上下文、12.8 万最大输出、2026 年 6 月知识截止，Haiku 系列首个支持可调 effort。定位是高频主力 + Opus/Sonnet 5.5 之下的 subagent。
How much does Claude Haiku 5.5 cost?Claude Haiku 5.5 怎么收费？ ▼
The headline rate is $0.10 input / $0.50 output per million tokens, but only for prompts up to 100,000 tokens. Cross 100K and the whole request reprices fivefold: $0.50 input / $2.50 output per million tokens. Anthropic says it is 90% cheaper than Haiku 4.5 below the threshold, 50% cheaper above it, and about 75% cheaper on a real workload. Confirm current pricing on the official site.$0.10 输入 / $0.50 输出每百万 token 的 headline 价，只适用于 100K 以内的 prompt。超过 100K 整单 5 倍跳涨：$0.50 输入 / $2.50 输出。Anthropic 称：100K 内比 Haiku 4.5 便宜 90%，100K 外便宜 50%，真实负载平均便宜约 75%。价格以官网为准。
What is the adjustable effort setting in Haiku 5.5?Haiku 5.5 的可调 effort 是什么？ ▼
Haiku 5.5 is the first Haiku-class model with adjustable effort: developers choose low, medium (default), high, xhigh or max per request, trading latency and cost against output quality. Anthropic's own docs suggest comparing the top two effort levels against Sonnet 5.5 before using them on hard work.Haiku 5.5 是 Haiku 系列首个支持可调 effort 的模型：每次调用选 low/medium（默认）/high/xhigh/max，在延迟、成本与质量之间取舍。官方文档建议，难任务用最高两档之前，先和 Sonnet 5.5 对比测一下。
Claude Haiku 5.5 vs Sonnet 5.5 — which should I use?Claude Haiku 5.5 和 Sonnet 5.5 怎么选？ ▼
Use Haiku 5.5 for high-volume, speed-sensitive work: summaries, classification, extraction, support chat, browser tasks and subagents. Keep Sonnet 5.5 as the default for open-ended coding: Anthropic's own figures show Sonnet 5.5 at high effort scoring 70.6% vs Haiku 5.5's 39.2% on Terminal-Bench 4.0 while costing about 45% less per task at max effort.高频、速度敏感的活（摘要、分类、抽取、客服、浏览器任务、subagent）用 Haiku 5.5；开放式编程 Sonnet 5.5 仍是默认：官方数据 Terminal-Bench 4.0 Sonnet 5.5（high effort）70.6% 对 Haiku 5.5 39.2%，且 max effort 下单任务成本还低约 45%。
Was this page helpful? 此页面对你有帮助吗？