Anthropic发布Claude Haiku 5.5：成本降75%且支持可调

Introducing Claude Haiku 5.5: the cheapest, fastest, and most capable small model we’ve ever released.

Claude Haiku 5.5 is designed for high-volume, cost-sensitive tasks. It reliably handles quick and repetitive workloads (like summaries, compactions, database queries, and classification requests). It pairs well with Opus 5.5 and Sonnet 5.5 as a subagent on coding work. And, since it’s also our fastest model to date, it works especially well for speed-sensitive tasks like live customer support and browser use.¹

Haiku 5.5 is available at a much lower price than Haiku 4.5. On average, it now costs around 75% less to run.²

Along with this launch, we’re making improvements to the value of our model range. We’re halving the price of Claude Sonnet 5.5’s cache reads, which means Sonnet 5.5 now runs around 20% cheaper on most agentic work. And we’re introducing a new monthly API credit for our Claude Max and Team subscribers, designed to support our users in building new agents and applications that run on the Claude Platform.

Performance

Here’s how Claude Haiku 5.5 performs across a range of benchmarks:

For details on how we run our evaluations, see the Haiku 5.5 System Card.

Haiku 5.5 is our first Haiku-class model to come with an adjustable effort setting. This means that, as with our other models, users can decide whether to optimize for cost or intelligence. The charts below show how Haiku 5.5 performs on three benchmarks at each effort setting:

Computer use: OSWorldKnowledge work: GDPval-AAMultidisciplinary reasoning: Humanity’s Last Exam

Computer use: OSWorldKnowledge work: GDPval-AAMultidisciplinary reasoning: Humanity’s Last Exam

OSWorld 2.1 (offline subset)Accuracy vs. cost

0102030405060708090Partial-credit score (%)0.050.100.200.50125Cost per attempt (USD, log scale)LowMedHighXhighMax

OSWorld 2.1 measures how well agents can operate a real computer to finish long, multi-step tasks.

GDPval-AA v2.1Accuracy vs. cost

8001000120014001600180020000Elo, as reported0.0050.010.020.050.100.200.50125Cost per task (USD, log scale)LowMedHighXhighMax

Artificial Analysis’s GDPval-AA v2.1 evaluates agents on real-world professional work across 44 occupations.

Humanity’s Last Exam (no tools)Accuracy vs. cost

010203040506070Score (%)0.0050.010.020.050.100.200.501Cost per attempt (USD, log scale)LowMedHighXhighMax

Humanity’s Last Exam (HLE) is a test of expert-level academic knowledge and reasoning.

In early testing, our customers reported results consistent with the performance and cost improvements shown above. Here’s what they told us about the new model:

AsanaHubSpotAlphaSenseBoxRogoCognition

AsanaHubSpotAlphaSenseBoxRogoCognition

Quote

“We’re very impressed with Claude Haiku 5.5, particularly its speed. We ran it through our eval suite for AI Teammates, our AI agent product, covering use cases like triaging bugs, setting up projects, and searching large portfolios to surface high-risk or overdue work. Compared with the model we use today, we saw over a 30% reduction in latency for task completions and up to 2.5x faster inference per agent turn. It’s a noticeably snappier experience.”

CompanyAsana

AuthorAaron Vinh, Staff Software Engineer

Quote

“At HubSpot, we use simulated portals to evaluate new models on CRM tasks like reporting on deals. We mostly test the smaller, more efficient models, and Claude Haiku 5.5 got the best score we’ve seen on this suite yet, at 92.8% averaged over three runs. One CRM audit task asks models to identify stale but ambiguous records. Across all of the models we tested, Haiku 5.5 was fastest to complete the task, and had the highest hit rate and the lowest false positive rate.”

CompanyHubSpot

AuthorZe’ev Klapow, Distinguished Software Engineer

Quote

“Ask in Document is one of our big sources of spend, doing about 8M calls a week in production. It answers very specific questions on top of one or a few documents. We ran 400 queries, and Claude Haiku 5.5 was a statistically significant improvement over Haiku 4.5: 0.84 vs. 0.76.”

CompanyAlphaSense

AuthorDaniel Campos, Distinguished Engineer

Quote

“Our customers use Box AI across large volumes of their enterprise content. With widespread usage comes the need to manage efficiency and cost, and to find the best model to suit the task at hand. In early testing, Claude Haiku 5.5 scored 11 points higher than Haiku 4.5 at about half the latency. We’d put it to use on analytical work that runs at scale, from cost reports to financial summaries and weekly recurring reviews.”

CompanyBox

AuthorYashodha Bhavnani, VP of AI Products

Quote

“The short and high-volume work is where Claude Haiku 5.5 fits for us, like quick lookups, subagents, and summaries. While a bigger model builds the deck, a Haiku 5.5 subagent goes into the 10-K and pulls the segment revenue line the deck needs. It’s accurate enough that we’d trust it there, and fast and cheap enough that we can run it a lot.”

CompanyRogo

AuthorAlex Wang, Applied AI

Quote

“Claude Haiku 5.5 joins the sidekick lineup in Devin Fusion as an excellent option. With Haiku 5.5 as the sidekick, Fusion holds a top-tier FrontierCode score of 66.2 while cutting cost and latency. You can try it today in the Devin CLI with Opus 5.5 as the lead.”

CompanyCognition

AuthorWalden Yan, Co-Founder & CPO

Pricing

The table below shows how Claude Haiku 5.5’s pricing compares to our other models. Haiku 5.5 is especially good value when used for tasks with prompts up to 100,000 tokens, which make up around 90% of requests to our previous Haiku model.

Safety

Alignment. Claude Haiku 5.5 shows major improvements across almost all of our alignment evaluations relative to Haiku 4.5. In particular, we found far fewer instances of misaligned behavior, and a lower willingness to cooperate with misuse. The model’s system card describes our evaluation process and results in more detail.

Safeguards. Consistent with its capabilities, Haiku 5.5’s cybersecurity safeguards are more restrictive than Haiku 4.5’s, but somewhat less restrictive than those we’ve applied to other recent models. In cybersecurity, they permit a wider range of defensive tasks than our safeguards for Sonnet 5.5, but they still block penetration testing and other techniques more likely to be used by attackers.