Anthropic has unveiled Claude Haiku 5.5, a lean model engineered to slash costs while boosting speed across high-volume applications. Designed for summaries, classification, database queries, and live customer support, Haiku 5.5 runs roughly 75% cheaper than its predecessor and introduces an adjustable effort setting to let developers trade off between cost and intelligence.
Two people collaborate at a desk in an Anthropic office, reviewing Claude Haiku 5.5 performance metrics on dual monitors.
Anthropic has released Claude Haiku 5.5, designed for high-volume, cost-sensitive tasks like summaries, data compaction, database queries, and classification requests. The model also serves as a subagent for coding work alongside Opus 5.5 and Sonnet 5.5, and excels at speed-sensitive applications including live customer support and browser automation.
Haiku 5.5 costs approximately 75% less to run than Haiku 4.5 on average.
The launch includes broader pricing improvements: Anthropic is halving the price of Claude Sonnet 5.5's cache reads—reducing Sonnet 5.5's cost on most agentic work by around 20%—and introducing a new monthly API credit for Claude Max and Team subscribers to support agent and application development on the Claude Platform.
Performance
Across benchmarks, Haiku 5.5 demonstrates substantial improvements over Haiku 4.5:
Benchmark
Haiku 5.5
Haiku 4.5
GPT-6 Luna
Sonnet 5.5
GDPval-AA v2.1
1620
735
1437
1840
AA-Briefcase v1.1
1578
614
1336
1824
OSWorld 2.1 (offline)
72.4%
15.7%
48.9%
83.9%
Humanity's Last Exam (no tools)
45.9%
10.2%
—
56.9%
Humanity's Last Exam (with tools)
57.4%
18.7%
—
64.5%
Terminal-Bench 4.0
39.2%
0.0%
16.4%
70.6%
FrontierCode 1.1 (Main)
46.4%
—
42.4%
52.1%
Chartography (no tools)
46.4%
6.4%
29.1%
61.6%
Haiku 5.5 is the first Haiku-class model to include an adjustable effort setting, allowing users to optimize for either cost or intelligence. OSWorld 2.1 measures how well agents operate computers to complete multi-step tasks. GDPval-AA v2.1 evaluates agents on real-world professional work across 44 occupations. Humanity's Last Exam tests expert-level academic knowledge and reasoning.
Early customer testing confirmed results consistent with the performance and cost improvements shown above.
Pricing
Haiku 5.5 offers particular value for tasks with prompts up to 100,000 tokens, which represent around 90% of requests to the previous Haiku model.
Price per 1 million tokens
Haiku 5.5 (≤100k / >100k tokens)
Haiku 4.5
Sonnet 5.5
Cache reads
$0.01 / $0.05
$0.10
$0.10
Cache writes
$0.125 / $0.625
$1.25
$2.50
Input tokens
$0.10 / $0.50
$1.00
$2.00
Output tokens
$0.50 / $2.50
$5.00
$10.00
Safety
Alignment. Claude Haiku 5.5 shows major improvements across nearly all alignment evaluations relative to Haiku 4.5, with substantially fewer instances of misaligned behavior and lower willingness to cooperate with misuse. Details are available in the model's system card.
Safeguards. Haiku 5.5's cybersecurity safeguards are more restrictive than Haiku 4.5's but somewhat less restrictive than other recent models. They permit a wider range of defensive tasks than Sonnet 5.5's safeguards while still blocking penetration testing and techniques more likely to be used by attackers.
Haiku 5.5's biology safeguards match those of Sonnet 5, Sonnet 5.5, and Opus 5, allowing research biology questions while restricting requests likely to cause harm. Organizations conducting broader biology and cyber activities can apply to Anthropic's Life Sciences Verification Program and Cyber Verification Program.
Availability and Additional Updates
Claude Haiku 5.5 is available now across all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure. On the Claude Platform, developers can access it via claude-haiku-5-5.
Anthropic is also making concurrent improvements to its broader model offerings:
Sonnet 5.5 price reduction. Cache reads for Claude Sonnet 5.5 now cost 50% less at $0.10 per million tokens (down from $0.20), reducing the cost of Sonnet 5.5 on most agentic tasks by approximately 20%.
API credits for subscribers. Starting this week, Max and Team subscribers on the Claude Platform will receive monthly API credits: Max 5x users get $100 per month, Max 20x users get $200, and Team subscribers receive up to $500 pooled across users. These credits support experimentation with tools, apps, and agents calling the API and can be used on any Anthropic model.
SDK updates. Anthropic is updating its Claude Python and TypeScript SDKs to add beta support for computer use and browser automation. Haiku 5.5 is well-suited to these tasks given its combination of speed, capability, and cost. More information is available in the Claude Platform docs.
—
Footnotes:
1 Claude Haiku 5.5 is the fastest model at each model's standard speed, though Opus models run faster in Fast Mode.
2 Haiku 5.5 pricing is 90% lower than Haiku 4.5 for requests up to 100,000 tokens and 50% lower for requests over 100,000 tokens. Haiku 4.5 saw 90% of requests fall into the former category. The comparison also accounts for tokenizer changes: Haiku 5.5 uses an updated tokenizer (matching Sonnet 5.5's and Opus 5.5's) that consumes slightly more tokens per task.
