Claude Haiku 4.5
Anthropic's smallest model, now a generation behind and only worth using as a subagent that searches large document collections.
Quality
Modality
multimodal
Context
200K tokens
Access
closed
Fabian's Take
"Haiku is completely outdated at this point. The only place it still makes sense for me is as a subagent that searches large document collections for a stronger model."
Claude Haiku 4.5 is the smallest model in the Claude family, optimized for low latency and high volume. It’s also the oldest one still in the current lineup.
How it compares
Haiku 4.5 dates from October 2025, and the rest of the lineup has moved on several times since. Sonnet 5 costs only twice as much per token, has a 1M-token context window instead of 200K, and is far stronger, so for most jobs, including the routine ones, it’s the better small model to hand work to.
The job Haiku still does well is reading. Point it at a large collection of documents as a subagent, let it search and pull out what’s relevant, and hand the result back to a stronger model like Opus to do the thinking. For that it’s fast and cheap enough to be worth keeping. Haiku 5.5 is announced for the coming weeks, which may change the picture.
The Verdict
Best for: A subagent that searches large document collections for a stronger model.
Pros
- Incredibly fast response times
- Very cheap API pricing
- Great for formatting and data extraction
Cons
- A generation behind the rest of the lineup, and it shows
- Struggles with complex reasoning
- Smaller context window than its siblings
Specs
- Pricing $1/M input, $5/M output
- Cost Tier budget
- ⚡Speed Tier instant
- License Proprietary
