Anthropic Releases Claude Haiku 5.5 With Effort Controls and Prices Up to 90% Lower

Anthropic on October 7, 2026 released Claude Haiku 5.5, its smallest model, priced at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. That is 90% below Claude Haiku 4.5 on those requests, and Haiku 5.5 is the first Haiku model with an adjustable effort setting, the company said.

Haiku 4.5 launched on October 15, 2025 at $1 and $5 per million tokens and stayed the newest Haiku for almost a year, according to Anthropic's documentation. Haiku 5.5 is Anthropic's third 5.5 model in about a month, after Opus 5.5 and Sonnet 5.5 (The New Stack), and it joins cheap small models from OpenAI and Google that compete mostly on price.

What did Anthropic release?

Claude Haiku 5.5 is a small model for high-volume work such as summaries, classification, database queries and subagent tasks, Anthropic said in its announcement on October 7, 2026. Anthropic calls it its fastest model to date at standard speed, though it runs slower than the Opus models in Fast Mode.

The model overview lists a 1 million token context window and up to 128,000 output tokens, up from 200,000 and 64,000 on Haiku 4.5. Effort sets how much the model thinks before it answers: Haiku 5.5 offers five levels from low to max, with medium as the default.

Anthropic made two other price changes on the same day. Cache reads on Claude Sonnet 5.5 drop from $0.20 to $0.10 per million tokens, which Anthropic says makes Sonnet 5.5 about 20% cheaper on most agentic work. Max 5x, Max 20x and Team subscribers also get monthly API credits of $100, $200 and up to $500, a recurring offer next to the one time $1,000 credit in the Claude Startups program that Anthropic expanded on October 6.

How much does Claude Haiku 5.5 cost?

Claude Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens, and $0.50 and $2.50 above that, according to Anthropic. Anthropic says about 90% of requests to Haiku 4.5 fell under that limit, and it puts the average saving at around 75%.

The saving per task is smaller than the per token cut. Anthropic's migration guide says the same text produces about 30% more tokens on Haiku 5.5 than on Haiku 4.5, because of a newer tokenizer.

Set side by side, the Anthropic and OpenAI price lists match to the cent on short prompts: GPT-6 Luna, OpenAI's current small model, also costs $0.10 for input, $0.50 for output, $0.01 for cached input and $0.125 for cache writes. The gap opens on long prompts. Anthropic's higher rate starts at 100,000 tokens and is five times the base price, while OpenAI charges double for input and 1.5 times for output only above 272,000 tokens. For a 200,000 token prompt, that works out to $0.50 per million input tokens on Haiku 5.5 against $0.10 on Luna. The two companies use different tokenizers, so equal prices per token do not mean equal prices per task.

Google lists Gemini 3.5 Flash-Lite at $0.30 and $2.50. Cheaper models exist: Alibaba's Qwen3.7 Flash costs $0.03 for input and $0.13 for output on prompts up to 32,000 tokens (The New Stack).

How does Haiku 5.5 compare with GPT-6 Luna and Haiku 4.5?

In Anthropic's own table, Claude Haiku 5.5 scores higher than OpenAI's GPT-6 Luna on 6 of 6 rows where both models have a score, and lower than Claude Sonnet 5.5 on 8 of 8 rows.

GPT-6 Luna is the newest Luna model on OpenAI's model list as of October 7, 2026, so the comparison is not against an outdated rival. The widest gap with Haiku 4.5 is on computer use, where an AI model operates a desktop or browser on its own: 72.4% against 15.7% on OSWorld 2.1.

Early customer results that Anthropic published, such as AlphaSense's 0.84 against 0.76 for Haiku 4.5 on 400 document questions, come from the customers themselves. Anthropic says Sonnet 5.5 and Opus 5.5 remain the better choice for complex agentic coding.

What changes for developers who switch?

Moving from Haiku 4.5 to Haiku 5.5 requires code changes, according to Anthropic's migration guide. Requests that set a fixed thinking budget, a non default temperature or a prefilled assistant reply now return an error, and computer use moves to a new toolset on the Claude API and Google Cloud.

Haiku 5.5 runs safety classifiers that can decline a request with no automatic fallback, and Priority Tier capacity is not supported. Its cyber safeguards are stricter than Haiku 4.5's but allow more defensive work than Sonnet 5.5's, while still blocking penetration testing, Anthropic says. Organizations that need more can apply to the Cyber Verification Program that Anthropic split into three tiers on October 6.

The model is available on the Claude Platform, Amazon Bedrock, Google Cloud and Microsoft Azure, and through gateways such as Vercel AI Gateway and OpenRouter.

What we don't know yet

• How Haiku 5.5 scores in independent tests, including on Artificial Analysis benchmarks run by Artificial Analysis itself.
• How much the average saving varies by workload once the larger token counts are included.
• Whether the tiered prices are identical on Amazon Bedrock, Google Cloud and Microsoft Azure.