claude-haiku-4-5 — pricing, context window and limits

The cheapest and fastest model in the current lineup

Haiku 4.5 is the current lineup's smallest and least expensive model per token, and the only current model with a 200K context window and a 64K output cap rather than the larger figures the rest of the lineup carries. It's the right default for high-volume, low-complexity work where speed and cost matter more than reasoning depth.

No effort parameter — this is the one to check for in shared code

Unlike every other current model, the effort parameter errors outright here rather than being accepted or ignored. Any request-building code shared across your whole model fleet needs to branch on this specifically. See the effort parameter errors on Haiku 4.5 for exactly this failure.

A cache minimum that catches short prompts off guard

This model's minimum cacheable prefix is among the highest in the current lineup, which means short prompts that cache cleanly elsewhere can silently fail to cache here at all, with no error to flag it. That combination — cheapest per-token rate, but a high bar to clear before caching even applies — is the detail most likely to make a Haiku migration underperform its price on paper.

Related

See migrating: Haiku 3 to Haiku 4.5 if you're moving off the older, deprecated Haiku generation, and Opus 5 vs Haiku 4.5 for how far apart the two ends of the current lineup actually sit.