Claude Haiku 4.5
Version: 20251001
Models from Partners and Community
These models constitute the vast majority of the Azure AI Foundry Models and are provided by trusted third-party organizations, partners, research labs, and community contributors. These models offer specialized and diverse AI capabilities, covering a wide array of scenarios, industries, and innovations. An example of models from Partners and community are the family of large language models developed by Anthropic. Anthropic includes Claude family of state-of-the-art large language models that support text and image input, text output, multilingual capabilities, and vision. See Anthropic's privacy policy to know more about privacy. Learn how to deploy Anthropic models . Characteristics of Models from Partners and Community:- Developed and supported by external partners and community contributors.
- Diverse range of specialized models catering to niche or broad use cases.
- Typically validated by providers themselves, with integration guidelines provided by Azure.
- Community-driven innovation and rapid availability of cutting-edge models.
- Standard Azure AI integration, with support and maintenance managed by the respective providers.
Key capabilities
About this model
Claude Haiku 4.5 delivers near-frontier performance for a wide range of use cases, and stands out as one of the best coding and agent models–with the right speed and cost to power free products and scaled sub-agents.Key model capabilities
- Extended thinking: Extended thinking gives Claude enhanced reasoning capabilities for complex tasks.
- Image & text input: Claude Haiku 4.5 can process images and return text outputs to analyze and understand charts, graphs, technical diagrams, reports, and other visual assets.
Use cases
See Responsible AI section for additional consideration for responsible use.Key use cases
- Powering free tier user experiences: Claude Haiku 4.5 delivers near-frontier performance at a cost and speed that makes powering free agent products and agentic use cases economically viable at scale.
- Real-time experiences: Claude Haiku 4.5's speed is ideal for real-time applications like customer service agents and chatbots where response time is critical.
- Coding sub-agents: Use Claude Haiku 4.5 to power sub-agents, enabling multi-agent systems that tackle complex refactors, migrations, and large feature builds with quality and speed.
- Financial sub-agents: Use Claude Haiku 4.5 to monitor thousands of data streams—tracking regulatory changes, market signals, and portfolio risks to preemptively adapt compliance and trading systems at previously impossible scales.
- Research sub-agents: Perform parallel analyses across multiple data sources while maintaining fast response times. Ideal for rapid business intelligence, competitive analysis, and real-time decision support.
- Business tasks: Claude Haiku 4.5 is capable of producing and editing office files like slides, documents, and spreadsheets. It also better supports strategy and campaign planning, business analysis and brainstorming.
Out of scope use cases
Please refer to the Claude Haiku 4.5 system card .Pricing
Pricing is based on a number of factors. See pricing details here .Technical specs
Please refer to the Claude Haiku 4.5 system card .Training cut-off date
July 2025Input formats
Image & text input: Claude Haiku 4.5 can process images and return text outputs to analyze and understand charts, graphs, technical diagrams, reports, and other visual assets. Text output: Claude Haiku 4.5 can output text of a variety of types and formats, such as prose, lists, Markdown tables, JSON, HTML, code in various programming languages, and more.Supported language
Claude Haiku 4.5 can understand and output a wide variety of languages, such as French, Standard Arabic, Mandarin Chinese, Japanese, Korean, Spanish, and Hindi. Performance will vary based on how well-resourced the language is.Sample JSON response
200:
{
"content": [
{
"text": "Hi! My name is Claude.",
"type": "text"
}
],
"id": "msg_213Zva2CMHLNnXjNJJKqJ2EG",
"model": "claude-haiku-4-5-20251001",
"role": "assistant",
"stop_reason": "end_turn",
"stop_sequence": null,
"type": "message",
"usage": {
"input_tokens": 31,
"cache_creation_input_tokens": 0,
"cache_read_input_tokens": 0,
"cache_creation": { "ephemeral_5m_input_tokens": 0, "ephemeral_1h_input_tokens": 0 },
"output_tokens": 25,
"service_tier": "standard",
}
}
4XX:
{
"error": {
"message": "Invalid request",
"type": "invalid_request_error"
},
"request_id": "<string>",
"type": "error"
}
Model architecture
Please refer to the Claude Haiku 4.5 system card .Optimizing model performance
Please refer to the Claude Haiku 4.5 system card .Additional assets
- Claude Documentation : Visit Anthropic's Claude documentation for a wealth of resources on model capabilities, prompting techniques, use case guidelines, and more.
- Extended Thinking Guide : Understand how best to use extended thinking with Claude.
- Claude Prompting Resources : Check out Anthropic's prompting tools and guides to learn how to craft prompts that elicit more helpful, nuanced responses.
- Claude Cookbooks : Check out example code for a variety of complex tasks, such as RAG from various web sources, making SQL queries, function calling, multimodal prompting, and more.
Distribution channels
- Claude API: For developers interested in building agents, Haiku 4.5 is available on the Claude Developer Platform.
- Claude Code: Use Haiku 4.5 with Anthropic's industry-leading coding agent, Claude Code.
More information
Data handling
By default, we may process customer data in select countries in the US, Europe, Asia and Australia. We will only store data in data centers located in the United States. For more on data handling and retention, see our Privacy Center.By default, we will not use your inputs or outputs from our commercial products (Anthropic API and Claude Code Enterprise) to train our models. If you explicitly report feedback or bugs to us or otherwise choose to allow us to use your data, then we may use your chats and coding sessions to train our models.
To find out more information regarding your use of an Anthropic commercial offering, or if you would like to know how to contact us regarding a privacy related topic, see our Trust Center and Commercial Terms.
Responsible AI considerations
Safety techniques
The Claude Haiku 4.5 system card describes in detail the evaluations Anthropic ran to assess the model's safety and alignment.Safety evaluations
Claude Haiku 4.5 shows large safety improvements compared to its predecessor, Claude Haiku 3.5. The new model's safety profile also compares favorably with other extant Anthropic models. The Claude Haiku 4.5 system card includes details of safety evaluations, including assessments of: the model's safeguards; the model's safety profile when working autonomously in “agentic” roles; the model's broad alignment; the model's own potential welfare; the model's tendency to “reward hack” by finding shortcuts to complete tests; and the model's potential to be misused to produce dangerous weapons.Known limitations
Please refer to the Claude Haiku 4.5 system card .Acceptable use
Acceptable use policy
Anthropic's Usage Policy is intended to help our users stay safe and promote the responsible use of our products and services.Quality and performance evaluations
Claude Haiku 4.5 delivers near-frontier performance for a wide range of use cases, and stands out as one of the best coding and agent models–with the right speed and cost to power free products and scaled sub-agents.| Benchmark | Test Name | Haiku 4.5 Score |
|---|---|---|
| Agentic coding | SWE-bench Verified | 73.3% |
| Agentic terminal coding | Terminal-bench | 41.0% |
| Agentic tool use | t2-bench | Retail 83.2%, Airline 63.6%, Telecom 83.0% |
| Computer use | OSWorld | 50.7% |
| High school math competition | AIME 2025 | 96.3% (python), 80.7% (no tools) |
| Graduate-level reasoning | GPQA Diamond | 73.0% |
| Multilingual Q&A | MMLU | 83.0% |
| Visual reasoning | MMMU (validation) | 73.2% |
Benchmarking methodology
SWE-bench Verified: All Claude results were reported using a simple scaffold with two tools—bash and file editing via string replacements. We report 73.3%, which was averaged over 50 trials, no test-time compute, 128K thinking budget, and default sampling parameters (temperature, top_p) on the full 500-problem SWE-bench Verified dataset. The score reported uses a minor prompt addition: "You should use tools as much as possible, ideally more than 100 times. You should also implement your own tests first before attempting the problem." Terminal-Bench: All scores reported use the default agent framework (Terminus 2), with XML parser, averaging 11 runs (6 without thinking (40.21% score), 5 with 32K thinking budget (41.75% score)) with n-attempts=1. τ2-bench: Scores were achieved averaging over 10 runs using extended thinking (128k thinking budget) and default sampling parameters (temperature, top_p) with tool use and a prompt addendum to the Airline and Telecom Agent Policy instructing Claude to better target its known failure modes when using the vanilla prompt. A prompt addendum was also added to the Telecom User prompt to avoid failure modes from the user ending the interaction incorrectly. AIME: Haiku 4.5 score reported as the average over 10 independent runs that each calculate pass@1 over 16 trials with default sampling parameters (temperature, top_p) and 128K thinking budget. OSWorld: All scores reported use the official OSWorld-Verified framework with 100 max steps, averaged across 4 runs with 128K total thinking budget and 2K thinking budget per-step configured. MMMLU: All scores reported are the average of 10 runs over 14 non-English languages with a 128K thinking budget. All other scores were averaged over 10 runs with default sampling parameters (temperature, top_p) and 128K thinking budget.Public data summary
N/AModel Specifications
Context Length200000
Quality Index0.84
Training DataJuly 2025
Last UpdatedJanuary 2026
Input TypeText,Image
Output TypeText
ProviderAnthropic
Languages8 Languages
Related Models