Claude Haiku 4.5: Anthropic's Fastest, Cheapest Model
Claude Haiku 4.5 is Anthropic's fastest and cheapest model. At $1 per million input tokens and $5 per million output tokens with a 200K context window, it is built for high-volume, latency-sensitive workloads where speed and unit economics matter more than peak reasoning. If you run production assistants, RAG pipelines, or tool-using agents at scale, Haiku is the model you reach for first.
What Claude Haiku 4.5 is best for
Haiku 4.5 is the volume workhorse in the Claude lineup. It is the fastest model Anthropic ships and the cheapest to run, which makes it the right default for any workload where you are paying per token at scale or where response latency is part of the product experience.
- High-volume API workloads where per-token cost dominates the bill
- Latency-sensitive user-facing features like chat, autocomplete, and live assistants
- Production RAG and retrieval pipelines that fan out across many calls
- Tool-using agents that make frequent, cheap model calls in a loop
- Classification, extraction, and routing tasks that do not need flagship reasoning
Pricing
Haiku 4.5 costs $1 per million input tokens and $5 per million output tokens. That is the lowest pricing in the Claude family. For comparison, Sonnet 4.6 runs $3 in / $15 out and Opus 4.8 runs $5 in / $25 out (with a Fast Mode at $10/$50). On top of that, batch processing is 50% cheaper and prompt caching cuts cached input by up to 90%, so the effective cost on repetitive or cache-friendly workloads drops well below the headline rate.
If your workload is high-volume or latency-bound, start on Haiku and only move up to Sonnet or Opus where the task actually demands it. Most production traffic does not.
Context window
Haiku 4.5 has a 200K token context window. That is smaller than the 1M token windows on Opus 4.8 and Sonnet 4.6, but 200K is still large enough for full documents, long conversations, and most RAG retrievals. If you genuinely need to load an entire large codebase or document set into a single prompt, step up to Sonnet 4.6 or Opus 4.8, both of which offer 1M tokens at flat rates with no surcharge.
How it compares
Within Claude, model choice should follow the workload. Sonnet 5 is now Anthropic's default on Claude Free and Pro and is available in Claude Code and the API, so use it as the current comparison baseline. Keep Haiku 4.5 in your evaluation set when testing latency-sensitive or high-volume steps. Like other Claude models, Haiku handles text and code, supports web search and image input, and works with MCP. It does not generate images, and voice is not a core differentiator.
Should you use it
Use Haiku 4.5 when speed and cost are the constraint and the task does not require flagship-level reasoning. It is ideal for production assistants, RAG, classification, and tool-using agents running at volume. Skip it when you need the largest context window, the deepest agentic coding, or native image generation. For those, move up to Sonnet 4.6 or Opus 4.8. Note that image generation is off the table across the entire Claude lineup, which stays text and code focused.
Specs at a Glance
| Spec | Claude Haiku 4.5 |
|---|---|
| Released | Claude Haiku 4.5 (fastest, cheapest model in the lineup) |
| Context window | 200K tokens |
| Input price (per M tokens) | $1 |
| Output price (per M tokens) | $5 |
| Best for | High-volume, latency-sensitive workloads, production assistants, RAG, tool-using agents |
| Image generation | No (text and code focused) |
| Web search | Yes |
Get the Claude playbook in your inbox.
One weekly email for Claude and Claude Code users. Real workflows, no hype. Subscribe and we send you The Claude Power-User Cheatsheet.