> ## Content Index
> Fetch the complete content index at: https://www.tao.media/llms.txt
> Use this file to discover other available public pages before exploring further.

# Anthropic Releases Claude Haiku 5.5 at Around 75% Lower Cost Than Haiku 4.5
- URL: https://www.tao.media/anthropic-releases-claude-haiku-5-5-at-around-75-lower-cost-than-haiku-4-5/
- Published: 2026-10-12T15:18:00.000Z
- Updated: 2026-10-12T15:21:41.000Z
- Description: Anthropic's new small model lowers average Haiku runtime costs, adds adjustable effort, and gives developers a cheaper subagent option for high-volume AI workflows.
- Author: Bart Hillerich
- Tags: Anthropic, AI, News

[Anthropic](https://www.anthropic.com/?ref=tao.media) has released Claude Haiku 5.5, a new small model the company says is its cheapest, fastest, and most capable Haiku model to date, oriented around high-volume agent work, summaries, classification, context compaction, and subagent calls.

The model is available now on the Claude Platform, Amazon Web Services, Google Cloud, and Microsoft Azure under the model ID `claude-haiku-5-5`.

The headline change is price. Anthropic says Claude Haiku 5.5 costs around 75% less to run on average than Claude Haiku 4.5, with the largest savings on shorter prompts. For prompts up to 100,000 tokens, Haiku 5.5 is priced at $0.10 per million input tokens and $0.50 per million output tokens, compared with $1.00 and $5.00 for Haiku 4.5.

That gives Anthropic a cheaper entry point for developers building agent systems that rely on many lightweight model calls under more expensive reasoning models such as Claude Sonnet and Claude Opus.

![Claude Haiku 5.5 benchmark table | Claude](https://pbs.twimg.com/media/HUC6-r6XoAAonZ5.png)

Claude Haiku 5.5 benchmark table | [Claude](https://x.com/claudeai/status/2107894042235166750?ref=tao.media)

## Haiku 5.5 Targets the High-Volume Layer of AI Agents

Small models increasingly handle the background work inside production AI systems. They summarize documents, route requests, classify data, compact context, query databases, and support larger models as subagents. Those calls can run constantly inside multi-step workflows, making price and latency just as important as peak benchmark performance.

Anthropic is pitching Haiku 5.5 directly at that layer. The company says the model can support customer support, browser automation, coding subagents, and other tasks where developers need low cost and fast responses across large request volumes.

Haiku 5.5 is also the first Haiku-class model with an adjustable effort setting. Developers can choose a lower-effort mode when cost and latency matter most, or raise effort when a task needs stronger reasoning. That gives Haiku 5.5 a feature previously associated with Anthropic's larger models and makes it easier to use the same small model across different levels of task difficulty.

![OSWorld effort-level chart | Claude](https://pbs.twimg.com/media/HUC7HfJWgAAqA_T.png)

OSWorld effort-level chart | [Claude](https://x.com/claudeai/status/2107894048753164362?ref=tao.media)

## Anthropic Reports Large Benchmark Gains Over Haiku 4.5

Anthropic's published benchmark table shows sizable improvements over Haiku 4.5 across knowledge work, computer use, coding, and visual reasoning. The company reports a score of 1620 on GDPval-AA v2.1, compared with 735 for Haiku 4.5 and 1437 for OpenAI's GPT-6 Luna. On OSWorld 2.1, which evaluates agents operating a computer, Haiku 5.5 reaches 72.4% on the offline subset, compared with 15.7% for Haiku 4.5 and 48.9% for GPT-6 Luna.

The coding results also move sharply higher. Haiku 5.5 scores 39.2% on Terminal-Bench 4.0, where Haiku 4.5 scored 0.0%, and 46.4% on FrontierCode 1.1\. On Humanity's Last Exam, Anthropic reports 45.9% without tools and 57.4% with tools, up from 10.2% and 18.7% for Haiku 4.5.

Anthropic still places Sonnet 5.5 above Haiku 5.5 on every listed benchmark, including 70.6% on Terminal-Bench 4.0 and 83.9% on OSWorld 2.1\. Sonnet and Opus remain the higher-end options for complex agentic coding and reasoning, while Haiku 5.5 is meant to make the cheaper, faster portion of the stack more capable.

All benchmark results are Anthropic-reported and have not yet been independently reproduced.

## Pricing Falls Most on Shorter Prompts

Haiku 5.5's pricing uses separate rates for prompts up to 100,000 tokens and prompts above that threshold. For prompts up to 100,000 tokens, cache reads cost $0.01 per million tokens, cache writes cost $0.125, input tokens cost $0.10, and output tokens cost $0.50\. Above 100,000 tokens, those prices rise to $0.05 for cache reads, $0.625 for cache writes, $0.50 for input tokens, and $2.50 for output tokens.

![Haiku 5.5 pricing table | Claude](https://pbs.twimg.com/media/HUC7MOWWkAAFnvw.png)

Haiku 5.5 pricing table | [Claude](https://x.com/claudeai/status/2107894052041490537?ref=tao.media)

Anthropic says the model is 90% cheaper than Haiku 4.5 for requests up to 100,000 tokens and 50% cheaper for requests above 100,000 tokens. The company said about 90% of Haiku 4.5 requests fell under the 100,000-token threshold. The average 75% savings figure also accounts for a tokenizer change in Haiku 5.5, which Anthropic says uses slightly more tokens per task.

The launch also includes pricing and developer-platform changes beyond Haiku. Anthropic said Sonnet 5.5 cache reads are being cut 50% to $0.10 per million tokens, which the company says makes Sonnet about 20% cheaper for most agentic work. Max and Team subscribers are also getting monthly API credits, with $100 for Max 5x, $200 for Max 20x, and up to $500 pooled for Team plans. Anthropic's Python and TypeScript SDKs now include beta support for computer use and browser use.

## Early Customers Report Lower Latency and Higher Scores

Anthropic included early customer feedback from Asana, HubSpot, AlphaSense, and Box. Asana Staff Software Engineer Aaron Vinh said the company saw "over a 30% reduction in latency for task completions and up to 2.5x faster inference per agent turn." HubSpot Distinguished Software Engineer Ze'ev Klapow said Haiku 5.5 produced "the best score we've seen on this suite yet, at 92.8%" on the company's CRM task benchmark.

AlphaSense Distinguished Engineer Daniel Campos said Haiku 5.5 produced a statistically significant improvement over Haiku 4.5, while Box VP of AI Products Yashodha Bhavnani said the model scored 11 points higher than Haiku 4.5 at about half the latency.

[Cursor](https://x.com/cursor%5Fai/status/2107897245282799864?ref=tao.media) added Haiku 5.5 on launch day and said the model costs 10x less than Claude Haiku 4.5 on shorter requests.

## What Haiku 5.5 Means for AI Agent Economics

Haiku 5.5 changes the cost structure beneath Anthropic's larger models instead of replacing them. In many agent systems, a small number of expensive reasoning calls sit on top of many cheaper support calls. If the small-model layer becomes cheaper and more capable, developers can run more subagents, more context processing, and more task-specific checks without raising total system cost as quickly.

AI companies, developer-tool startups, and enterprise teams are trying to move agents from demos into production. Lower per-task costs can make browser use, coding assistants, customer-support workflows, and internal knowledge agents easier to operate at scale.

The launch also puts more pressure on open-weight and decentralized inference providers. Haiku 5.5 lowers the proprietary-model price floor for short, high-volume tasks while adding stronger computer-use and coding results than Anthropic's prior small model. Competing providers will need to beat not only the raw token price, but the cost per completed task once model quality, latency, caching, and orchestration are included.

Anthropic says Haiku 5.5 also improves across almost all of its alignment evaluations compared with Haiku 4.5, with cybersecurity safeguards that are tighter than Haiku 4.5's but less restrictive than Sonnet 5.5's for defensive work.

Haiku 5.5 gives Anthropic a cheaper small-model tier for the background work that makes agent systems run. The larger Claude models still lead the company's benchmark table, but Haiku 5.5 lowers the cost of using them inside broader, multi-model workflows.