Claude Haiku 4.5 is a newly released model from Anthropic, offering performance comparable to Sonnet 4. Its pricing enables large-scale deployment of intelligent solutions, and while Haiku 4.5’s performance is close to that of Sonnet 4.5, it costs only about one-third as much!
Claude Haiku 4.5 is now available on DepolyAI!
Users can access the latest AI model directly on the DepolyAI website — no VPN, no installation, and pricing is fully aligned with the official rates.
Go try it out below!
https://www.deployai365.com/
Next, let’s take a look at the new features of Claude Haiku 4.5!
Claude Haiku 4.5 supports a 200K-token context window and can generate up to 64K tokens in output. The model is capable of processing both text and images, making it Anthropic’s first Haiku model to include extended reasoning, computer use, and context awareness capabilities.
The extended reasoning and computer use features have been available in the Sonnet and Opus models for several months, but context awareness is a relatively new innovation.
In benchmark testing, Haiku 4.5 achieved an accuracy of 73.3% on SWE-bench Verified, a test based on real GitHub issues from open-source projects.
By comparison, Sonnet 4.5 scored 77.2% on the same benchmark.
This means that Haiku 4.5’s accuracy is within five percentage points of the current best-in-class models, while costing only one-third as much ($1 / $5 vs. $3 / $15 per million input/output tokens).
In the OSWorld computer use benchmark, Haiku achieved an accuracy of 50.7%, the highest score ever recorded by a Haiku model on this test.
We’ll now compare the differences between Haiku 3.5 and Haiku 4.5:
Haiku 3.5 (Previous Generation):
- No deep reasoning
- No computer use
- No context awareness
- $0.80 / $4 per million tokens
Haiku 4.5 (Current Version):
- Extended reasoning (explains reasoning before responding)
- Computer use (takes screenshots, clicks, fills out forms)
- Context awareness (model knows its own token consumption)
- $1 / $5 per million tokens
Claude Haiku 4.5 Pricing
Claude Haiku 4.5 is priced at $1 per million input tokens and $5 per million output tokens, representing a 25% increase over Haiku 3.5’s $0.80 / $4 pricing.
For batch processing, output tokens receive a 50% discount, lowering the effective rate to $1 / $2.50.
Instant caching costs $1.25 per million tokens written and $0.10 per million tokens read.
However, compared to the Claude Sonnet 4 model, Claude Haiku 4.5 now offers similar coding performance at one-third the cost and is more than twice as fast.
Claude Haiku 4.5 Applications
With Claude Haiku 4.5, users can control applications through screenshots and actions.
The model can analyze a screenshot of the current screen, decide what to do (e.g., click a button, type text, scroll down, or navigate to a URL), and execute that action.
On the OSWorld benchmark, Haiku 4.5 achieved a 50.7% success rate, representing its performance under real-world test conditions.
In terms of context, OSWorld evaluates models with tasks such as “Find the quarterly revenue figure in this financial dashboard” or “Fill out this vendor onboarding form using the provided information.”
These are realistic automation scenarios where a system must interact with software without an API or when the API doesn’t expose necessary functionality.
A 50.7% success rate indicates that Haiku 4.5 successfully completed about half of such tasks under benchmark conditions.
At present, computer use is best suited for manual workflows where errors are manageable.
It’s not designed for mission-critical real-time operations, but it’s ideal for repetitive, click-heavy tasks that might otherwise take a human hours to complete.
A 50.7% success rate, combined with appropriate supervision, can deliver meaningful practical value.
If you’re interested, you can try the latest Claude Haiku 4.5 model now on the DepolyAI official website below!