The cost of high-volume AI just dropped again. On October 7, 2026, Anthropic launched Claude Haiku 5.5, which the company describes as its fastest, cheapest and most capable small model yet. For short-context requests, the price falls by roughly 90% compared with Haiku 4.5. For an SME that classifies emails, summarizes documents, or runs a support chatbot all day, this drop directly changes the budget equation for automation.
In brief
- Anthropic launched Claude Haiku 5.5 on October 7, 2026, two weeks after Claude Sonnet 5.5.
- For requests under 100,000 context tokens, pricing drops to $0.10 per million input tokens and $0.50 per million output tokens, versus $1 and $5 for Haiku 4.5, a roughly 90% cut at that tier.
- Factoring in real mixed usage, Anthropic reports an average saving of 75% versus Haiku 4.5.
- On the OSWorld 2.1 computer-use benchmark, the score jumps from 15.7% to 72.4% between generations.
- It is the first Haiku model to offer an adjustable effort dial, letting teams trade off cost against intelligence per task.
A model built for volume, not for the hardest problems
Claude Haiku 5.5 sits at the entry point of the Claude lineup, below Sonnet 5.5 and Opus 5.5. Its job isn't to compete on the hardest reasoning tasks, but to handle a very large number of simple tasks at the lowest possible cost: classifying emails or tickets, short summaries, database lookups, first-line customer support, or subtasks delegated by a larger AI agent.
Anthropic explicitly recommends this model for high-volume, low-complexity-per-task scenarios, which is exactly the kind of work an SME tends to automate first: sorting a sales inbox, pre-qualifying inbound leads, or generating standardized meeting summaries.
Quick definition: the effort dial
Claude Haiku 5.5 is the first Haiku model to offer several adjustable effort levels. A company can ask for a fast, cheap answer for a simple task, or request deeper reasoning (at higher cost) for a trickier case, without switching models.
A price cut that scales by tier
The most concrete part of the announcement is the pricing grid, which works in tiers based on context length.
| Tier | Claude Haiku 4.5 | Claude Haiku 5.5 |
|---|---|---|
| Input (up to 100,000 tokens) | $1 / million tokens | $0.10 / million tokens |
| Output (up to 100,000 tokens) | $5 / million tokens | $0.50 / million tokens |
| Input (beyond 100,000 tokens) | $1 / million tokens | $0.50 / million tokens |
| Output (beyond 100,000 tokens) | $5 / million tokens | $2.50 / million tokens |
Source: Anthropic, Claude Haiku 5.5 launch announcement, October 7, 2026; figures cross-checked with coverage from The New Stack and PPC Land.
Anthropic notes that the cheapest tier covers about 90% of the traffic observed on the previous Haiku 4.5 model, meaning most real-world usage benefits from the maximum price cut automatically, with no extra configuration.
Clear gains on agent and browsing tasks
A cheaper model only matters if it stays capable. Anthropic published several benchmark results comparing Haiku 5.5 with Haiku 4.5, and, as a reference point, with OpenAI's competing GPT-6 Luna model.
On OSWorld 2.1, a benchmark that measures an AI agent's ability to use a computer the way a human would (navigating, clicking, filling in fields), the score jumps from 15.7% to 72.4%. On Terminal-Bench 4.0, a technical agentic benchmark, Haiku 5.5 reaches 39.2%, ahead of GPT-6 Luna according to Anthropic, but well behind Sonnet 5.5 (70.6%), which confirms its role as a support model rather than a flagship for the most demanding tasks.
Claude Haiku 5.5
Claude Sonnet 5.5
Why this price cut matters for SMEs
For an SME leader, the benchmark numbers matter less than what they unlock financially. A model that is 90% cheaper on short tasks makes automations profitable that weren't a few months ago: a chatbot answering hundreds of questions a day, automated invoice sorting, or applicant screening in recruiting.
The same caution applies as with any model switch: test before rolling out widely. A cheaper model that produces lower-quality answers on a specific use case can end up costing more in human corrections than the previous model, even with a lower sticker price. The cost-versus-quality trade-off should always be measured against a real task, not a listed price.
Identify high-volume tasks
Test Haiku 5.5 on a sample
Roll out with a safety net
FAQ
What is Claude Haiku 5.5?
Claude Haiku 5.5 is the smallest, cheapest AI model in Anthropic's Claude lineup, launched on October 7, 2026. It is designed for simple tasks handled at very high volume: classification, summaries, customer support, data lookups.
How much does Claude Haiku 5.5 cost?
For requests with context under 100,000 tokens, pricing is $0.10 per million input tokens and $0.50 per million output tokens. Beyond that threshold, rates rise to $0.50 and $2.50. These figures represent roughly a 90% cut versus Haiku 4.5 on the first tier.
Can Claude Haiku 5.5 replace a more powerful model like Sonnet 5.5?
Not for every task. Haiku 5.5 still trails on complex reasoning and demanding agentic benchmarks (39.2% versus 70.6% for Sonnet 5.5 on Terminal-Bench 4.0). It suits simple, high-volume tasks, not work that requires deep reasoning.
Where is Claude Haiku 5.5 available?
The model is available on Anthropic's Claude Platform, as well as on Amazon Web Services, Google Cloud and Microsoft Azure, under the technical identifier claude-haiku-5-5.
Automating with the right model at the right cost is now a real trade-off for SMEs to manage. To go further, see our guide to choosing an AI model or explore all our AI resources for business leaders.
