The Lab Notes

Anthropic

Claude Haiku 5.5 costs 90% less than Haiku 4.5 on short prompts

Anthropic released Claude Haiku 5.5 with input priced at $0.10 per million tokens for prompts up to 100,000 tokens.

The Lab Notes ·

A large origami crane folded from flat panels of coral red, blue, lime green and pale peach with black outlines, facing right on a dark navy background with thin white arcs and lines on the left, while a repeating pattern of smaller identical cranes fills the right side.

Anthropic released Claude Haiku 5.5, a small model priced 90% below Haiku 4.5 for requests up to 100,000 tokens. Anthropic calls it “the cheapest, fastest, and most capable small model we’ve ever released.”

Anthropic announced the model Oct. 7, 2026, in its Introducing Claude Haiku 5.5 post. The model ID is claude-haiku-5-5. It is generally available, not a preview, on Amazon Web Services, Google Cloud and Microsoft Azure. Anthropic’s Haiku product page, dated Oct. 7, says Claude.ai Free, Pro, Max, Team and Enterprise plans can select it and that it is in Claude Code.

Anthropic said it is designed for high-volume, cost-sensitive tasks such as summaries, database queries and classification. It pairs with Opus 5.5 and Sonnet 5.5 as a subagent, a helper model that takes smaller jobs inside a larger coding task.

Input costs $0.10 per million tokens, a tenth of Haiku 4.5

Anthropic’s pricing page, read Oct. 7, 2026, lists Haiku 5.5 at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. Above that length, input is $0.50 and output is $2.50. Haiku 4.5 costs $1.00 for input and $5.00 for output.

A cache read is a charge for reusing text the model has already processed. Haiku 5.5 charges $0.01 per million tokens up to 100,000 and $0.05 above, against $0.10 on Haiku 4.5.

Anthropic said the cut is 90% for requests up to 100,000 tokens and 50% for longer ones. On Haiku 4.5, 90% of requests fell in the shorter group. An updated tokenizer, the tool that splits text into the units that get billed, uses slightly more tokens per task. After that, Anthropic said, Haiku 5.5 costs around 75% less to run on average.

Haiku 5.5 is the first Haiku-class model with an adjustable effort setting, Anthropic said.

Anthropic’s benchmarks put Haiku 5.5 far ahead of Haiku 4.5

Anthropic’s Oct. 7 post reports these figures, which are its own, for Haiku 5.5, Haiku 4.5 and Sonnet 5.5 in that order.

  • GDPval-AA v2.1, an Artificial Analysis eval scored in Elo, where higher is better: 1620, 735 and 1840. GPT-6 Luna scored 1437.
  • Terminal-Bench 4.0: 39.2%, 0.0% and 70.6%. GPT-6 Luna scored 16.4%.

Anthropic said Sonnet 5.5 and Opus 5.5 remain better choices for complex agentic coding like Terminal-Bench 4.0. It said Haiku 5.5 is best suited to narrowly scoped tasks such as compaction, summarization or subagent work.

Asana and HubSpot report faster and more accurate runs

Asana’s Aaron Vinh said, as quoted in Anthropic’s post, that tests against the model Asana uses today showed over a 30% reduction in latency for task completions. HubSpot’s Ze’ev Klapow said Haiku 5.5 scored 92.8% averaged over three runs on its simulated-CRM-portal suite, the best score HubSpot has seen on that suite.

Anthropic halved Sonnet 5.5 cache reads the same day

Starting Oct. 7, Anthropic cut Sonnet 5.5 cache reads 50%, to $0.10 per million tokens from $0.20. It said this lowers Sonnet 5.5’s cost on most agentic tasks by around 20%. Sonnet 5.5 input is $2.00 per million tokens.

Haiku 5.5 still blocks penetration testing

Anthropic said Haiku 5.5’s cybersecurity safeguards permit a wider range of defensive tasks than Sonnet 5.5’s but still block penetration testing.

Analysis

We think the price cut matters most on short prompts. For prompts up to 100,000 tokens, Haiku 5.5 input at $0.10 is one-tenth of Haiku 4.5 and one-twentieth of Sonnet 5.5’s $2.00. Anthropic’s own account points the model at narrow work. Its Terminal-Bench 4.0 figure of 39.2% against Sonnet 5.5’s 70.6% fits that advice for hard coding.