What happened
On October 7, 2026, Anthropic released Claude Haiku 5.5, the newest model in its smallest and cheapest tier. It is available through the Claude Platform API as claude-haiku-5-5 and on Amazon Bedrock, Google Cloud Vertex AI and Microsoft Azure. Anthropic positions it for high-volume, cost-sensitive work and as a fast subagent next to Opus 5.5 and Sonnet 5.5.
What’s new
Pricing is split at 100k prompt tokens. Per million tokens:
| Haiku 5.5 (up to 100k / over 100k) | Haiku 4.5 | |
|---|---|---|
| Input | $0.10 / $0.50 | $1.00 |
| Output | $0.50 / $2.50 | $5.00 |
| Cache reads | $0.01 / $0.05 | $0.10 |
| Cache writes | $0.125 / $0.625 | $1.25 |
Anthropic says that works out to roughly 75% lower cost on average: 90% cheaper than Haiku 4.5 for requests up to 100k tokens and 50% cheaper above that. It notes that Haiku 5.5 uses a newer tokenizer, so the same task consumes slightly more tokens.
- Effort control: the first Haiku-class model with an adjustable effort setting (Low through Max).
- Benchmarks (vendor-reported): OSWorld 2.1 offline subset 72.4% (Haiku 4.5: 15.7%); Terminal-Bench 4.0 39.2% (Haiku 4.5: 0.0%); Humanity’s Last Exam 45.9% without tools and 57.4% with tools.
- Speed: Anthropic calls it its fastest model at standard speed.
- Safety: cybersecurity safeguards are stricter than Haiku 4.5’s and looser than Sonnet 5.5’s; penetration testing and attacker-oriented techniques remain blocked.
In the same announcement Anthropic halved Sonnet 5.5 cache-read prices to $0.10 per million tokens and added beta computer-use and browser-use support to its Python and TypeScript SDKs.
Why it matters
Small models handle most of the token volume in production systems: classification, summarization, context compaction, extraction and routing. A tenfold price drop at that tier changes the economics of agent pipelines that call a cheap model many times per task. Anthropic itself says Sonnet 5.5 and Opus 5.5 remain the better choice for complex agentic coding, so Haiku 5.5 is best treated as a worker or subagent, not a replacement for the larger models.
Teams currently on Haiku 4.5 should re-benchmark: the price per token is lower, but the new tokenizer means per-task cost will not fall by exactly the headline percentage.
How to try it
- Change the model string to claude-haiku-5-5 in existing Messages API calls; Anthropic links a migration guide from the launch page.
- Start with a low effort level for classification or extraction, and raise it only where quality drops.
- Keep prompts under 100k tokens where possible to stay in the lower price band.