1. Home
2. News
3. Anthropic

Claude Haiku 5.5 cuts prices up to 90% and tops GPT-6 Luna in Anthropic's own tests

Most requests now cost a tenth of what Haiku 4.5 charged, and the small model gets an effort dial for the first time.

By Think Facility · October 7, 2026 · Anthropic Products Save

On October 7, 2026, Anthropic released Claude Haiku 5.5, the new version of its smallest model. It's built for high-volume jobs like summaries and classification, and it's out now on the Claude Platform and in Claude Code, as well as on AWS, Google Cloud and Microsoft Azure.

The price is the news. On prompts up to 100,000 tokens, Haiku 5.5 charges a tenth of what Haiku 4.5 did, and Anthropic says prompts that size made up about 90% of requests to the old model.

Input tokens

$0.10 per 1 million up to 100k, $0.50 over

Output tokens

$0.50 per 1 million up to 100k, $2.50 over

Haiku 4.5

$1.00 input, $5.00 output

Model ID

claude-haiku-5-5

Why the average saving is 75%

Above 100,000 tokens the cut is 50%, so Anthropic puts the typical saving at around 75% rather than 90%. There's a second reason it gives in a footnote. Haiku 5.5 has a new tokenizer, the part that chops text into the units you're billed for, and it counts a bit higher.

It's also the first Haiku with an effort control, so you can choose between cheaper answers and more reasoning on each call, as you already could with Sonnet and Opus.

How it stacks up against GPT-6 Luna

Anthropic's chart puts Haiku 5.5 beside OpenAI's GPT-6 Luna, the model that starts powering free ChatGPT on October 8. On the six tests where both have a score, Haiku 5.5 comes out ahead on all of them. The widest gaps are on computer use, 72.4% against 48.9% on OSWorld 2.1, and on Terminal-Bench 4.0, a command-line coding test, 39.2% against 16.4%.

Those are Anthropic's own runs, so I'd wait for outside numbers before treating them as settled. And Haiku 5.5 doesn't catch the bigger Claude models. Sonnet 5.5 scores 70.6% on that same coding test, and Anthropic says Sonnet and Opus remain the better choice for complex agentic coding.

Anthropic's Claude account led with the cost when it posted the launch at 2:01 pm Eastern.

What early testers said

Six companies that tried it early are quoted in the announcement. AlphaSense, which runs about 8 million calls a week through one document feature, gave the most concrete comparison.

On safety, Anthropic says Haiku 5.5's cyber safeguards are stricter than Haiku 4.5's but looser than those on its other recent models. They still block penetration testing.

Cheaper Sonnet, and credits for Max and Team

Anthropic made two other changes the same day. Cache reads on Sonnet 5.5 drop from $0.20 to $0.10 per million tokens, which it says makes Sonnet 5.5 about 20% cheaper on most agentic work.

Max and Team subscribers also get a monthly API credit for the Claude Platform: $100 on Max 5x, $200 on Max 20x, and up to $500 pooled across a Team. Anthropic says the credits roll out this week and work on any of its models.