Vibeleaderboard
← All Intel
Intel / post

Claude Haiku 5.5 launches at about 75% lower run cost

Source
x.com
Date
claudeai@claudeai

Introducing Claude Haiku 5.5: the cheapest, fastest, and most capable small model we’ve ever released. On average, it costs around 75% less to run than Claude Haiku 4.5.

Context

Anthropic says Claude Haiku 5.5 costs around 75 percent less to run than Claude Haiku 4.5, but that figure is an average, and what a given user saves depends on how long their prompts are. Anthropic announced the model on October 7, 2026 as its cheapest, fastest and most capable small model. It is aimed at high-volume, cost-sensitive work such as summaries, classification and database queries, and at running as a , a helper model that a larger model such as Sonnet 5.5 or Opus 5.5 hands narrow jobs to during coding work.

The announcement page shows where the average comes from. Prices are set per million tokens, the chunks of text a model reads and writes. For requests up to 100,000 tokens, Haiku 5.5 is priced 90 percent below Haiku 4.5: $0.10 per million input tokens against $1.00, and $0.50 per million output tokens against $5.00. Above 100,000 tokens the discount falls to 50 percent. Anthropic says 90 percent of Haiku 4.5 requests fell under that line, and its calculation also accounts for a new tokenizer, the component that splits text into tokens, which makes Haiku 5.5 use slightly more tokens per task. The 75 percent average reflects that mix of mostly short requests. Requests over 100,000 tokens, such as very long documents or large amounts of code, get only the 50 percent discount, trimmed further by the extra tokens, so a workload with more of them than Haiku 4.5's typical mix should save less than the average.

Anthropic does not present Haiku 5.5 as a stand-in for its larger models. Its benchmark table shows large gains over Haiku 4.5 but a clear gap to Sonnet 5.5 on hard agentic work. On -Bench 4.0, which measures complex multi-step tasks in a , Haiku 5.5 scores 39.2 percent, Sonnet 5.5 scores 70.6 percent, and Haiku 4.5 scored 0.0 percent. Anthropic says Sonnet 5.5 and Opus 5.5 remain the better choices for complex agentic coding, and that Haiku 5.5 suits narrowly scoped tasks such as summarization and subagent work. Haiku 5.5 is also the first Haiku with an adjustable effort setting, which lets a developer choose whether to favor lower cost or more intelligence. All of these scores and savings are Anthropic's own figures. The evidence reviewed here includes no independent testing, and the customer results quoted on the page are early-testing reports published by Anthropic.

For work that still needs the larger model, the same announcement cuts its price too. Cache reads on Sonnet 5.5 now cost $0.10 per million tokens instead of $0.20. A cache read is billed when a model reuses text it has already processed, such as the same instructions or files sent again on every step of a long task. Anthropic says cache reads make up a large share of token use, so the change lowers Sonnet 5.5's cost on most agentic tasks by around 20 percent.

Terms in this piece · Glossary
  • multi-agent — Using several AI agents on one problem — splitting work in parallel, checking each other, or filling different roles like planner and reviewer.
  • Terminal — A program that shows text and accepts typed text.
  • Command-line interface — The names, options and inputs a program accepts when typed to it, and the text it prints back.
  • Prompt — Information or instructions supplied to a model; in a chat, the request you type is often called your prompt.
More from claudeai
Recommended reads
Comments

Checking sign-in…

Loading comments…