Claude Haiku 5.5 Price Cut: What '75% Cheaper' Really Means

In short: Anthropic released Claude Haiku 5.5 on 7 October 2026 at a much lower price than Haiku 4.5. Anthropic says it costs around 75% less to run on average. That figure is an average, and the real saving for your workload depends on how long your prompts are and on a tokenizer change. Test it on your own tasks before you switch.

Key Takeaways

• Prompts up to 100,000 tokens cost $0.10 per million input tokens and $0.50 per million output tokens. Longer prompts cost $0.50 and $2.50.
• Haiku 4.5 cost $1 and $5, so the headline drop is steep for short prompts.
• Anthropic says the model costs "around 75% less to run" than Haiku 4.5 on average. Reports do not explain how that average is calculated.
• A new tokenizer produces roughly 30% more tokens for the same text, which offsets part of the price cut.
• Anthropic positions the model for high-volume, latency-sensitive work such as classification, routing, extraction and subagent tasks.

What Did Anthropic Announce?

According to AI Weekly's report, Anthropic released Claude Haiku 5.5 on 7 October 2026 and says it costs "around 75% less to run on average than its predecessor, Haiku 4.5." The model is available on AWS, Google Cloud and Microsoft Azure, with the model ID claude-haiku-5-5.

VKTR's coverage adds the price table, a larger context window and the intended uses. It also notes that Anthropic will keep the model available until at least 7 October 2027.

What Does It Cost?

Per million tokens, according to VKTR:

The Batch API's 50% discount still applies. Haiku 5.5 is also the only current Claude model priced by prompt length.

Why Is It Not Simply "90% Cheaper"?

Some headlines say 90%. VKTR's author explains why that is the short-prompt best case, not Anthropic's own number:

• Input and output prices fall by about 90% for prompts up to 100,000 tokens.
• A new tokenizer produces about 30% more tokens for identical input, which gives back part of the saving.
• Long prompts above 100,000 tokens see a smaller saving, because their rates are higher.

VKTR's author estimates short-prompt workloads should cost roughly 85% to 90% less after the tokenizer effect, while Anthropic's own figure is a 75% average. Both are estimates. Your bill depends on your mix of prompt lengths, output lengths and caching.

What Is the Model Meant For?

Anthropic says it targets high-volume, latency-sensitive work such as classification, routing, extraction and subagent tasks. That fits tasks where a small, fast model does a narrow job many times a day, such as sorting support tickets or pulling fields from documents.

Anthropic's reported benchmarks for Haiku 5.5 against Haiku 4.5 are large. For example, AI Weekly lists 72.4% against 15.7% on an offline computer-use test. These are vendor-reported results, and customers quoted in the same report, such as Asana and HubSpot, describe their own tests that have not been independently verified.

What Should You Check Before Switching?

VKTR lists several changes that can break existing code:

• Thinking is on by default. An effort setting defaults to medium, and thinking tokens count toward max_tokens. A low limit can end a response before any text appears.
• Responses can begin with a thinking block. Code that reads the first content block as the answer may fail.
• Some sampling settings return errors. Non-default temperature, top_p or top_k values are rejected.
• Message arrays must end with a user turn, and the computer use tool version changed.
• Safety classifiers can decline requests with a refusal stop reason.

What Should a Business Do?

1. Run your own cost test. Replay a sample of real requests through both models and compare the total bill, not the price per million tokens.
2. Measure quality on your tasks. A cheaper model that gets 3% more answers wrong can cost more once review time is counted.
3. Route by task. Use the small model for classification, routing and extraction, and keep a larger model for hard reasoning.
4. Keep your system model-flexible. Prices and models change every few months, so avoid hard-wiring one provider.
5. Watch the migration list above before you change the model name in production.

The Takeaway

A lower price list is good news for anyone running AI at volume, and it widens the set of tasks where AI makes economic sense. But "75% cheaper" is an average from the vendor, not a promise about your bill. Measure it on your own work.

Frequently Asked Questions

How much does Claude Haiku 5.5 cost?

For prompts up to 100,000 tokens it costs $0.10 per million input tokens and $0.50 per million output tokens. For longer prompts it costs $0.50 and $2.50.

Is Claude Haiku 5.5 really 75% cheaper than Haiku 4.5?

Anthropic says it costs around 75% less to run on average. The real saving depends on your prompt lengths and on a tokenizer that produces about 30% more tokens for the same text.

What is Claude Haiku 5.5 good for?

Anthropic says high-volume, latency-sensitive work such as classification, routing, extraction and subagent tasks.

Do I need to change my code to use it?

Possibly. Reports note that thinking is on by default, some sampling settings return errors and responses can begin with a thinking block, so test before switching.

Where is Claude Haiku 5.5 available?

On the Claude API, and through AWS, Google Cloud and Microsoft Azure, according to the reports.

Related Insights

• Gemini 4 Argon Release: What Was Claimed, What Is Unproven
• AI Coding Agents: What Microsoft's Copilot Rethink Means

Sources

• AI Weekly, Anthropic releases Claude Haiku 5.5, cutting prices by 75%, 7 October 2026.
• VKTR, coverage of the Claude Haiku 5.5 launch and pricing, October 2026.