# Claude Haiku 5.5 Is Now Available Anthropic released Claude Haiku 5.5 at $0.10/$0.50 per million tokens, around 90% cheaper than Haiku 4.5, with effort levels and a 1M-token context window. It is now the fastest Claude option in AI Data Stream and powers our background tasks on Anthropic providers. Haiku 4.5 stays available as the previous generation. Anthropic released **Claude Haiku 5.5** yesterday, and it is now in the AI Data Stream model picker as the fastest Claude option. It is also the model we now use for the small background jobs that run on your Anthropic provider, so most Anthropic users get cheaper, faster titles and summaries without changing anything. Haiku 4.5 remains available as the previous generation. ## Pricing Haiku 5.5 is priced by prompt length on the standard API we use. For prompts up to 100,000 tokens, which covers almost everything AI Data Stream sends: - **Input:** $0.10 per million tokens - **Output:** $0.50 per million tokens - **Cache reads:** $0.01 per million tokens - **Cache writes:** $0.125 per million tokens Prompts over 100,000 tokens pay $0.50 input and $2.50 output per million tokens. Against Haiku 4.5 at $1/$5, that is a 90% cut for prompts up to 100,000 tokens and a 50% cut above. One caveat: Haiku 5.5 uses the newer tokenizer shared by Claude 4.7 and later, which counts roughly 30% more tokens for the same text, so the saving per request is a little smaller than the headline per-token figure. It is still by far the cheapest model in the picker. ## What's New This is the first Haiku of the 5.5 generation, and the gap to Haiku 4.5 shows in the numbers. On GDPval-AA, Artificial Analysis's benchmark of real professional work across 44 occupations, Haiku 5.5 scores 1620 Elo against 735 for Haiku 4.5, within reach of Sonnet 5.5's 1840 at a twentieth of the price. On the OSWorld computer-use benchmark it reaches 72.4% against 15.7% for Haiku 4.5. Asana reported "over a 30% reduction in latency for task completions and up to 2.5x faster inference per agent turn" in its testing. See [Anthropic's announcement](https://www.anthropic.com/claude-haiku-5-5) for the full results. Two things make it a different kind of Haiku in AI Data Stream: - **A 1M-token context window.** Haiku 4.5 stopped at 200K. Long conversations with lots of tool results no longer need a bigger model just for the room. - **Adaptive thinking with effort levels.** Haiku 4.5 had no effort control at all. Haiku 5.5 decides how much to think on each request and you steer it with the effort picker, so it can be a serious analysis model at High and a near-instant responder at Low. ## What We Changed Behind the Scenes AI Data Stream runs several small jobs on the fastest model of your provider: naming new conversations, summarising long annotation descriptions, summarising knowledge-base entries, and drafting system prompts in the AI prompt generator. For Anthropic providers, all of these now run on Haiku 5.5. In our testing it answered those prompts without spending any tokens on thinking, so they are now roughly a tenth of their previous cost and quick as before. ## Effort Levels Haiku 5.5 supports the full effort scale: **Low, Medium, High, Extra High, Max**. Anthropic's default for it is Medium, and the picker still starts on the effort you saved for your Anthropic provider. For quick questions, Low is the natural setting; for a longer analysis you can push it up and see how far a model this cheap gets. Haiku 4.5 has no effort control, so the picker stays hidden when it is selected. ## Haiku 4.5 Stays Available Haiku 4.5 is still in the picker, labelled as the previous generation, so you can compare the two on your own data. Anthropic has not announced a retirement date for it. If it is your default model, nothing changes until you switch. ## The Full Claude Lineup - **Claude Fable 5.1** — Most powerful, premium; for the hardest reasoning - **Claude Fable 5** — Previous-generation Fable - **Claude Opus 5.5** — Most capable at standard pricing, for demanding tasks - **Claude Opus 5** — Previous-generation Opus - **Claude Sonnet 5.5** — Best balance of speed and capability (default) - **Claude Sonnet 5** — Previous generation - **Claude Haiku 5.5** — Fastest and cheapest, great for quick queries and high-volume work - **Claude Haiku 4.5** — Previous-generation Haiku ## Switching Models 1. Go to **Settings > AI Providers** 2. Click your Anthropic provider 3. Select your preferred model from the dropdown 4. Save Or select it from the model picker when starting a new conversation. ## Questions? If you have any questions, [contact us](/contact/) or check our [AI providers documentation](/docs/ai-providers/).