Claude Haiku 5.5 slashes prices up to 90%, fueling AI pricing war

Huge price cuts, big speed jumps

Anthropic has released Claude Haiku 5.5, its fastest and most affordable small AI model yet. The company says the new version costs about 75% less than its predecessor on average, with some workloads dropping up to 90% ¹. That means developers can run many more tasks for the same money—a sign that the race to make AI cheaper is far from over.

The model is built for high-volume, cost-sensitive jobs like summarizing documents, answering database queries, and handling live customer-support chats. It’s also much better at “computer use”—tasks where the AI controls a screen or terminal to get work done—thanks to a leap in benchmark scores ².

Two pricing tiers, one catch

Haiku 5.5 uses a two-tier pricing system based on how long your prompt (the instruction you give the AI) is. For prompts up to 100,000 tokens—a token is a tiny chunk of text, roughly a word—the price is $0.10 per million input tokens and $0.50 per million output tokens ³. That’s a 90% cut compared to Haiku 4.5.

If your prompt goes over that 100k-token limit, the rate jumps fivefold to $0.50 in and $2.50 out. Even then it’s still half the old price, but the headline-grabbing 90% discount only applies to shorter requests ³. Anthropic says most earlier Haiku requests were under that threshold, so the real-world savings land closer to the 75% figure the company quotes.

The model also comes with an adjustable “effort” setting, letting users trade off between cost and intelligence. Think of it like choosing between economy and premium fuel for a car—you pay more when you ask for more reasoning power.

Beating rivals, but with a catch

On official benchmarks, Haiku 5.5 beats the previous Haiku model by a wide margin and leads OpenAI’s budget model, GPT-6 Luna, in every category where both are listed ¹. It scores 72.4% on a computer-use test (up from 15.7%), and 45.9% on a multidisciplinary reasoning exam without any tools (up from 10.2%) ².

Independent ranking site Artificial Analysis puts Haiku 5.5 at the top of its small-model league with an intelligence score of 43, just ahead of GPT-6 Luna’s 38 ¹. But there’s a trade-off: Haiku 5.5 burns through about three times as many output tokens as GPT-6 Luna to reach a similar score. It also makes up false information (a problem called “hallucination”) in 40% of cases, compared to 77% for Luna—though it’s better at admitting when it doesn’t know something .

What it means for the industry

Anthropic is also cutting the cost of reading cached data for its pricier Claude Sonnet 5.5 model in half, making agent-based coding tasks about 20% cheaper ². The company is adding monthly API credits for its Claude Max and Team subscribers, giving them extra room to experiment with new AI applications .

With the new model now available on AWS, Google Cloud, and Azure, the message is clear: making AI cheap and fast is a major battleground. As more companies release lower-cost models, the pricing war will likely keep driving innovation—and keeping bills down for developers.

Source markers

¹ The Decoder, October 8 2026 ² Anthropic launch post, October 7 2026 ³ AIToolsRecap breakdown, October 2026

Key facts

• Claude Haiku 5.5 costs up to 90% less than Haiku 4.5 for short prompts.
• It scores 72.4% on computer‑use tasks, a huge jump from 15.7%.
• Artificial Analysis ranks it the top small‑class model (score 43).
• The model uses an adjustable effort setting to balance cost and quality.
• Sonnet 5.5 cache‑read prices are cut in half alongside the launch.

Sources

1. Claude Haiku 5.5 arrives with massive price cuts proving the AI pricing arms race is far from over — the-decoder.com
2. Introducing Claude Haiku 5.5 \ Anthropic
3. Claude Haiku 5.5 Pricing: 90% Cut, Two Tiers | AIToolsRecap — aitoolsrecap.com
4. Claude Haiku 5.5: Features, Benchmarks, and Pricing — datacamp.com
5. Claude Haiku 5.5: Benchmarks, Pricing and Specs (2026 ... — theairankings.com