Anthropic's cheapest, fastest small Claude — tuned for high-volume, speed-sensitive work at $0.10 per million input tokens, with an adjustable effort setting.
Last verified · 2026-10-07 · by Moe Ameen
Claude Haiku 5.5 is the small, fast tier of Anthropic's Claude family, released on October 7, 2026 as the third model in the Claude 5.5 line after Opus 5.5 and Sonnet 5.5. Anthropic describes it as "the cheapest, fastest, and most capable small model we've ever released." The job of a Haiku model is speed and cost at volume — not frontier reasoning — and 5.5 pushes both further than any Haiku before it.
Pricing is the standout. Haiku 5.5 is tiered at a 100,000-token prompt boundary: up to 100k tokens it is $0.10 per million input tokens and $0.50 per million output; above 100k it rises to $0.50 input and $2.50 output, with cache reads from $0.01 per million. Against Haiku 4.5 ($1 input / $5 output) that is roughly a 90% cut on short prompts, 50% on long, and about 75% cheaper on average. It is also the first Haiku-class model with an adjustable effort setting — Low, Medium, High, Xhigh, Max — so you can dial depth up or down per call.
You reach it as `claude-haiku-5-5` on the Claude Platform API and across Amazon Web Services, Google Cloud, and Microsoft Azure. Anthropic positions it for high-volume, cost-sensitive work (summaries, classification, extraction, quick lookups), speed-sensitive work (live support, browser use), and as a fast subagent paired with Opus 5.5 or Sonnet 5.5 on coding. It is candid about the trade: on hard agentic benchmarks like Terminal-Bench 4.0 it trails Sonnet 5.5 by a wide margin, so it is the model for many fast, well-scoped jobs rather than one slow, hard one.
For a creator, the honest framing is that Haiku 5.5 is a drafting and processing engine, not a content tool. It takes text and image input and returns text — it generates no images, video, or audio, and it publishes nothing. Its value to content work is throughput: lots of cheap, quick language operations, fast.
Haiku 5.5's distinct edge is not depth, it is throughput: it drafts fast and costs almost nothing per call, which makes it a genuine ideation multiplier. Where a slower, pricier model nudges you toward one careful attempt, Haiku lets you generate forty hook variants, ten angle lists off a transcript, and a dozen caption options and then pick the winners. The problem is that forty good hooks are still just text in a window — none of them are a video, a carousel, or a scheduled post, and Haiku has no way to become one. That is the exact seam [Kompozy](/) is built to close.
Concretely: use Haiku 5.5 to brainstorm wide and cheap, then bring the handful of winning lines into Kompozy and watch each become finished media. A hook becomes a [Persona Short](/glossary/persona-shorts) delivered on camera by a face-locked avatar; a list of points becomes a brand-exact [HyperFrames](/glossary/hyperframes) carousel; a standout line becomes a Quote Graphic; the underlying idea becomes a Blog Article and an Email Newsletter — 18 formats across video, image, and text, every one held to your voice by the [Persona Brief](/glossary/persona-brief). Because Kompozy runs Claude generation itself, the polished copy stays Claude-class even as the cheap ideation happens in Haiku. Then [Autopilot](/glossary/autopilot) reframes and captions each asset for 9:16, 1:1, and 16:9 and schedules the set across eight social platforms plus blog and email behind a per-post review. Haiku 5.5 is how you explore cheaply; Kompozy is how the winners get built and shipped.
Claude Haiku 5.5 is the small, fast, cheap tier of Anthropic's Claude family, released October 7, 2026 as the third model in the Claude 5.5 line. Anthropic calls it its cheapest, fastest, most capable small model. It takes text and image input and returns text, tuned for high-volume and speed-sensitive work — it does not generate images, video, or audio.
It is tiered at a 100,000-token prompt boundary: $0.10 per million input tokens and $0.50 per million output up to 100k, and $0.50/$2.50 above, with cache reads from $0.01 per million. Anthropic says that is about 90% cheaper than Haiku 4.5 for short prompts, 50% for long, and roughly 75% cheaper on average. Check Anthropic's pricing page for current figures.
For hard, open-ended work, no — Anthropic says Sonnet 5.5 and Opus 5.5 remain the better choices for complex agentic tasks, and Haiku trails them on benchmarks like Terminal-Bench 4.0. Haiku 5.5 shines on many fast, well-scoped, high-volume jobs and as a subagent. The common pattern is routing cheap mechanical steps to Haiku and escalating to Sonnet or Opus when a task needs judgment.
No. It is a text and reasoning model that returns text only — no images, video, audio, design, or publishing. To turn its fast drafts into finished, scheduled posts you pair it with a content engine like Kompozy, which renders the media and publishes across eight social platforms plus blog and email.
Not necessarily. You can call claude-haiku-5-5 directly via the Claude API, but a managed engine like Kompozy already runs Claude generation for you and manages model choice internally — you write a Persona Brief and approve outputs rather than wiring models yourself.