[{"data":1,"prerenderedAt":14},["ShallowReactive",2],{"artikel-en-berita-claude-haiku-5-5":3},{"slug":4,"lang":5,"title":6,"description":7,"date":8,"updated":9,"topic":10,"product":9,"cover":11,"readMin":12,"html":13},"berita-claude-haiku-5-5","en","Claude Haiku 5.5: Anthropic's small model that costs 90% less","Anthropic released Claude Haiku 5.5 on October 7, 2026. A 1 million token context, 128,000 token output, and prices up to 90% cheaper than Haiku 4.5 for requests under 100,000 tokens.","2026-10-19",null,"berita","\u002Fimages\u002Fartikel\u002Fberita-claude-haiku-5-5.webp",4,"\u003Cp>Mid-October brought Claude Haiku 5.5, the model Anthropic calls its cheapest, fastest, and most capable Haiku-class release to date. Launch date: October 7, 2026. Average running cost is down around 75% versus Haiku 4.5, and for prompts under 100,000 tokens the drop reaches 90%.\u003C\u002Fp>\n\u003Cp>This is not just a price cut. It signals a shift in how people should pick AI models: stop defaulting to the most expensive one, and route work by weight.\u003C\u002Fp>\n\u003Ch2>1. What changed in Haiku 5.5\u003C\u002Fh2>\n\u003Cp>Three numbers to know:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>1 million token context.\u003C\u002Fstrong> Five times the 200,000 of Haiku 4.5. A single chat session can absorb thousands of pages of text, code, or documents without losing context.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>128,000 token output.\u003C\u002Fstrong> Up from 64,000. Useful when you need the model to write a long report or produce a large code block in one reply.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Adjustable "effort" parameter.\u003C\u002Fstrong> A first for the Haiku class. You choose whether the model reasons deeply (smarter, more expensive) or goes fast and cheap.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Inputs supported: text, images, video, and audio, matching Sonnet 5.5 and Opus 5.5.\u003C\u002Fp>\n\u003Ch2>2. Real pricing versus Haiku 4.5\u003C\u002Fh2>\n\u003Cp>Anthropic publishes two price tiers based on prompt size. For requests under 100,000 tokens (about 90% of requests to the previous Haiku model):\u003C\u002Fp>\n\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth>Per 1 million tokens\u003C\u002Fth>\n\u003Cth>Haiku 5.5\u003C\u002Fth>\n\u003Cth>Haiku 4.5\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\u003Ctr>\n\u003Ctd>Input\u003C\u002Ftd>\n\u003Ctd>$0.10\u003C\u002Ftd>\n\u003Ctd>$1.00\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>Output\u003C\u002Ftd>\n\u003Ctd>$0.50\u003C\u002Ftd>\n\u003Ctd>$5.00\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>Cache read\u003C\u002Ftd>\n\u003Ctd>$0.01\u003C\u002Ftd>\n\u003Ctd>$0.10\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\u003C\u002Ftable>\n\u003Cp>For requests over 100,000 tokens, prices rise to $0.50 input and $2.50 output. That is still 50% cheaper than Haiku 4.5.\u003C\u002Fp>\n\u003Cp>The implication: if most of your work is short (summaries, classification, quick answers), almost every request falls into the cheap tier. That is exactly the scenario Haiku 5.5 was built for.\u003C\u002Fp>\n\u003Ch2>3. Where you can use it\u003C\u002Fh2>\n\u003Cp>Haiku 5.5 is available on every major platform:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>Direct Claude API, model ID \u003Ccode>claude-haiku-5-5\u003C\u002Fcode>\u003C\u002Fli>\n\u003Cli>Amazon Bedrock\u003C\u002Fli>\n\u003Cli>Google Cloud\u003C\u002Fli>\n\u003Cli>Microsoft Azure\u003C\u002Fli>\n\u003Cli>Claude Code and GitHub Copilot from day one\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>If you already build on Claude or any of these clouds, switching is a model ID change. No major migration work.\u003C\u002Fp>\n\u003Ch2>4. Side benefit: Sonnet 5.5 gets cheaper too\u003C\u002Fh2>\n\u003Cp>Alongside this launch, Anthropic halved Sonnet 5.5's cache read price, from $0.20 to $0.10 per 1 million tokens. This matters for agentic work that relies heavily on caching, because Sonnet 5.5 now runs about 20% cheaper on most agent workloads.\u003C\u002Fp>\n\u003Cp>In addition, Max and Team subscribers now receive monthly API credits to use on the Claude Platform: $100 per month for Max 5x, $200 for Max 20x, and up to $500 for Team (pooled across members).\u003C\u002Fp>\n\u003Ch2>5. What it does not do\u003C\u002Fh2>\n\u003Cp>Haiku 5.5 is not a replacement for Opus 5.5 or Sonnet 5.5 on heavy work. Anthropic itself positions it as a subagent partner: the model that handles the lightweight steps in a larger workflow, so Opus or Sonnet can focus on the parts that need deeper reasoning.\u003C\u002Fp>\n\u003Cp>It is also not a frontier research model. Anthropic still keeps its internal research models under restricted access; what launched here is the production version.\u003C\u002Fp>\n\u003Ch2>What it means for you\u003C\u002Fh2>\n\u003Cp>This news reshapes an old question. "Which model is the smartest?" is giving way to "Which model fits this job?"\u003C\u002Fp>\n\u003Cp>If you run a business or build products on AI, four things are worth thinking about:\u003C\u002Fp>\n\u003Col>\n\u003Cli>\u003Cstrong>When to use Haiku 5.5.\u003C\u002Fstrong> For summaries, classification, database queries, and fast customer support answers. For these tasks it is enough, and far cheaper.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>When not to.\u003C\u002Fstrong> For complex analysis, long-form creative writing, or code that requires deep reasoning chains, stay on Sonnet 5.5 or Opus 5.5.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Combine models.\u003C\u002Fstrong> The most effective pattern: let Sonnet or Opus plan and decompose tasks, then let Haiku 5.5 do the repetitive parts. This is the "subagent" strategy Anthropic itself recommends.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Re-estimate your costs.\u003C\u002Fstrong> If you have an application that calls the Claude API heavily, moving the light steps to Haiku 5.5 can cut costs sharply. Recalculate before switching.\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>For everyday users who only use Claude chat: you will not see Haiku 5.5 directly. But it will power much of the background work that keeps Claude products fast and affordable to run.\u003C\u002Fp>\n\u003Cp>Want to know which AI tools suit your work right now? See my picks on the \u003Ca href=\"\u002Fen\u002Fai-tools\">AI Tools\u003C\u002Fa> page.\u003C\u002Fp>\n\u003Ch2>Sources\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fwww.anthropic.com\u002Fclaude-haiku-5-5\" target=\"_blank\" rel=\"noopener\">Introducing Claude Haiku 5.5\u003C\u002Fa> (Anthropic)\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fsupport.claude.com\u002Fen\u002Farticles\u002F12138966-release-notes\" target=\"_blank\" rel=\"noopener\">Claude Release Notes (October 2026)\u003C\u002Fa> (Anthropic Help Center)\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fdocs.anthropic.com\u002Fen\u002Frelease-notes\u002Fapi\" target=\"_blank\" rel=\"noopener\">Claude Platform API release notes\u003C\u002Fa> (Anthropic Docs)\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Famdatalakehouse.substack.com\u002Fp\u002Fhaiku-55-muse-wonder-open-racks-oct\" target=\"_blank\" rel=\"noopener\">Haiku 5.5, Muse Wonder, Open Racks: Oct 6-14\u003C\u002Fa> (Data, Lakehouse and AI)\u003C\u002Fli>\n\u003Cli>\u003Ca href=\"https:\u002F\u002Fnote.com\u002Fakiakisoftball\u002Fn\u002Fnc25186f8751b?hl=en\" target=\"_blank\" rel=\"noopener\">This Week's AI News Summary: October 9, 2026\u003C\u002Fa> (note.com)\u003C\u002Fli>\n\u003C\u002Ful>\n",1793543302423]