Anthropic 发布 Claude Haiku 5.5：1M 上下文，输入每百万 token 0.10 美元
Anthropic Releases Claude Haiku 5.5: A Small Model With 1M Context Priced at $0.10 per Million Input Tokens - MarkTechPost
Anthropic 发布 Claude Haiku 5.5，这是其目前最便宜最快的小模型，保持 1M token 上下文窗口和最高 128K 输出 token，100K token 以内提示的输入价格为每百万 token 0.10 美元、输出 0.50 美元，比 Claude Haiku 4.5 低 90%。该模型已通过 Claude API、Amazon Bedrock、Google Cloud、Microsoft Foundry 和 AWS 上的 Claude Platform 正式提供，是首个带可调 effort 设置的 Haiku 级模型，自适应思考默认开启、effort 默认 medium。Anthropic 自报的基准中，OSWorld 2.1 为 72.4%，Terminal-Bench 4.0 为 39.2%，Humanity's Last Exam 无工具 45.9%、带工具 57.4%。
推荐理由
AI 评分 80
Haiku 5.5 以 1M 上下文和大幅降价切入高频子智能体场景，读者可据此判断小模型的分工位置。
主题
正文
当前显示原文，中文全文译文尚未就绪。
Anthropic has released Claude Haiku 5.5, its cheapest and fastest small model to date. It targets high-volume work like summaries, compaction, classification and subagent tasks. It keeps a 1M token context window and up to 128K output tokens. Pricing starts at $0.10 per million input tokens and $0.50 per million output tokens. That is 90% below Claude Haiku 4.5 for prompts up to 100K tokens.
Is it deployable? Yes, as a hosted API. Haiku 5.5 is generally available on the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS.
What Anthropic Shipped
Haiku 5.5 is the first Haiku-class model with an adjustable effort setting. Adaptive thinking is on by default, and the effort parameter defaults to medium.
It takes text and images, outputs text, and has a June 2026 knowledge cutoff. Batch jobs support up to 300K output tokens in beta.
It is important to note two key things. Non-default temperature, top_p or top_k values return a 400 error. The new tokenizer also counts the same text as roughly 30% more tokens than Haiku 4.5. The migration guide covers both.
Pricing: A 2-Tier Structure
Pricing splits at 100K prompt tokens. Up to 100K, input costs $0.10 and output $0.50 per million. Cache reads cost $0.01 and 5-minute cache writes $0.125. Above 100K, rates rise to $0.50 input and $2.50 output.
GPT-6 Luna lists identical short-context rates. Its higher tier starts only above 272K input tokens, at $0.20 and $0.75. For a 150K-token prompt, Luna is cheaper on list price.
Benchmarks
All figures below are Anthropic-reported (see the system card):
OSWorld 2.1 (offline subset): 72.4%, versus 48.9% for GPT-6 Luna and 15.7% for Haiku 4.5; Terminal-Bench 4.0: 39.2%, versus 16.4% for Luna and 0.0% for Haiku 4.5; FrontierCode 1.1 (Main): 46.4%, versus 42.4% for Luna; Humanity’s Last Exam: 45.9% without tools and 57.4% with tools; GDPval-AA v2.1: 1620, versus 1437 for Luna and 735 for Haiku 4.5
Best Use Cases
Three workloads fit Haiku 5.5 best:
The first is subagent work under Opus 5.5 or Sonnet 5.5. At Rogo, a Haiku 5.5 subagent pulls a 10-K revenue line while a bigger model builds the deck; The second is high-volume document Q&A and summarization. AlphaSense tested it on a feature handling about 8M calls a week; The third is speed-sensitive work like live customer support and browser use
Haiku 5.5 vs Its Closest Competitors
Sources: Anthropic docs, Anthropic announcement, OpenAI model page, OpenAI pricing, Google model page, Google pricing. Standard-tier list prices, verified October 7, 2026.
Interactive Explainer
Key Takeaways
$0.10 input and $0.50 output per 1M tokens for prompts up to 100K tokens; 1M context, 128K output, adaptive thinking with an effort setting (default medium); 72.4% on OSWorld 2.1 versus 15.7% for Haiku 4.5 (Anthropic-reported); Same short-context list price as GPT-6 Luna; Luna is cheaper above 100K tokens; Positioned as a subagent under Opus 5.5 and Sonnet 5.5, not a coding lead
[Sponsored] The web is the one API most agents are missing. Databases, calendars and repos have APIs. The open web mostly doesn’t. The TinyFish MCP server gives any MCP client four tools: TinySearch, TinyFetch (full pages as markdown, JavaScript included), TinyBrowser for logins and forms, and TinyAgent for multi-step jobs. Search and Fetch are free.
Michal Sutter
Michal Sutter is a data science professional with a Master of Science in Data Science from the University of Padova. With a solid foundation in statistical analysis, machine learning, and data engineering, Michal excels at transforming complex datasets into actionable insights.
Michal SutterReka Releases Rho-1: A 19B Omni-Reasoning Model That Understands, Generates Video and Outputs Robot Actions in One; Michal SutterCan an Open Model Do Security Research? Cantina’s apex-flash-1 Solves 40 of 60 Held-Out Bug Tasks; Michal SutterGoogle Research Moves Federated Learning Into TEEs: Gboard Now Trains With Externally Verifiable Differential Privacy; Michal SutterDeepSeek Harness v0.2 Brings Official Desktop Apps to Its Open-Source Agent Harness