Technology & AI
Andrey Shcherbina
Dec 19, 2025
·
Updated on
Dec 19, 2025
On October 15, 2025, Anthropic released Claude Haiku 4.5—fastest and most affordable model in Claude lineup. What was frontier technology five months ago (Claude Sonnet 4) now available three times cheaper and more than twice as fast.
Haiku 4.5 showed 73.3% on the SWE-bench Verified test—almost at the level of Sonnet 4, which was a state-of-the-art model in spring 2025. Meanwhile, the price is only $1 per million input tokens and $5 per million output—three times cheaper than Sonnet 4.5.
What Is Claude Haiku 4.5
Claude Haiku 4.5 is a small language model from Anthropic, optimized for tasks where speed and cost matter. This isn't just a "lightweight" version of large models—Haiku 4.5 surpasses Sonnet 4 on some tasks, especially in computer control.
Model perfectly suited for real-time tasks: chat assistants, customer support, pair programming, rapid prototyping. Claude Code users will notice significantly more responsive experience when working with multi-agent projects.
Haiku 4.5 available through Anthropic API (identifier claude-haiku-4-5), claude.ai web application, iOS and Android mobile apps, and through cloud platforms Amazon Bedrock and Google Vertex AI. Context window is 200,000 tokens, maximum output up to 64,000 tokens, knowledge base current through February 2025.
Main Haiku 4.5 Achievements
Result of 73.3% on SWE-bench Verified—almost at level of Sonnet 4 (~60%), which was top model in spring 2025. Considering Haiku 4.5 is three times cheaper and twice as fast, this is an impressive achievement.
On some tasks, Haiku 4.5 surpasses Sonnet 4. Particularly in computer use, the model shows better results, making applications like Claude for Chrome faster and more useful.
First Haiku model with extended thinking support—mode when model "thinks" before answering. This improves the quality of solving complex tasks while maintaining speed advantages.
Security — Best Among All Models
By automated alignment assessment metric, Claude Haiku 4.5 showed statistically significantly lower levels of problematic behavior than Sonnet 4.5 and Opus 4.1. This makes Haiku 4.5 the safest Anthropic model at release time.
The model shows low risks in producing chemical, biological, radiological and nuclear weapons (CBRN). For this reason, it was released under ASL-2 safety level—less strict than ASL-3 for Sonnet 4.5 and Opus 4.1.
New Ways to Use Models Together
Haiku 4.5 opens new patterns for working with Claude models. For example, Sonnet 4.5 can break down complex problems into multi-step plans, then organize a team of several Haiku 4.5 models that execute subtasks in parallel.
This is especially useful for large projects: one expensive model plans, several cheap and fast models execute work. Result—money savings while maintaining high quality.
Comparison with Competitors
By October 2025, the small language model market became very competitive. Fast and cheap models appeared from all major players. Key comparison parameters—ratio of price, speed and quality.
For real-time tasks, not only accuracy matters but also latency. The model must respond instantly, otherwise the user experience suffers. Cost is also critical—with millions of requests daily, even a small price difference becomes significant.
Feature | Claude Haiku 4.5 | Claude Sonnet 4.5 | GPT-5.1 Mini | GPT-4o mini | Gemini 3 Flash |
Release date | Oct 15, 2025 | Sep 29, 2025 | Nov 2025 | Jul 2024 | Oct 2025 |
Developer | Anthropic | Anthropic | OpenAI | OpenAI | Google DeepMind |
Context | 200K tokens | 200K / 1 M (beta) | 128K tokens | 128K tokens | 1M tokens |
Max output | 64K tokens | 64K tokens | 32K tokens | 16K tokens | 8K tokens |
Price in/out | $1 / $5 per 1M | $3 / $15 per 1M | $0.40 / $1.60 per 1M | $0.15 / $0.60 per 1M | $0.35 / $1.05 per 1M |
SWE-bench Verified | 73.3% | 77.2% | ~65% | ~50% | ~55% |
Programming | Excellent | Excellent | Good | Average | Good |
Computer use | Excellent (better than Sonnet 4) | Excellent | Average | Weak | Average |
Speed | Very fast | Fast | Very fast | Very fast | Very fast |
Extended thinking | ✅ Yes | ✅ Yes | ❌ No | ❌ No | ❌ No |
Security | ASL-2 (best) | ASL-3 | Standard | Standard | Standard |
Multimodal | Text + images | Text + images | Text + images | Text + images | Text + images + video |
Availability | Everywhere | Everywhere | API, Chat GPT | API, Chat GPT | API, Gemini |
Comparison Conclusions:
Claude Haiku 4.5 offers best performance in the fast model category with 73.3% result on SWE-bench. This is 8-18% higher than competitors (GPT-5.1 Mini ~65%, Gemini 3 Flash ~55%, GPT-4o mini ~50%). Unique extended thinking support and excellent computer control make it the best choice for quality tasks.
Pricing and Usage
Claude Haiku 4.5 cost is $1 per million input tokens and $5 per million output tokens. This is three times cheaper than Sonnet 4.5 ($3/$15) and five times cheaper than Opus 4.5 ($5/$25).
For comparison with the previous version: Haiku 3.5 cost $0.80/$4, meaning Haiku 4.5 is 25% more expensive. However, performance gain significantly exceeds price increase—model approaches level of Sonnet 4, which cost $3/$15.
Additional savings available through prompt caching (up to 90% discount) and batch processing (50% discount). When using caching, cost can drop to $0.10/$0.50 per million tokens for repeated queries.
Access Through Applications
Haiku 4.5 available free to all Claude users—this is the default model for free tier. In claude.ai web application and iOS/Android mobile apps, you can select Haiku 4.5 in the model selector.
Model efficiency means users can complete more tasks within usage limits while maintaining premium model performance.
New Claude Haiku 4.5 Features for Developers
Haiku 4.5 available through Claude API, Amazon Bedrock and Google Vertex AI. The model serves as a direct replacement for both Haiku 3.5 and Sonnet 4—simply change model identifier to claude-haiku-4-5 and get better performance at most economical price.
For high-load projects, it is recommended to use Haiku 4.5 for subtasks not requiring maximum accuracy, and Sonnet 4.5 for mission-critical operations. This pattern optimizes quality-cost balance.
Claude Code with Haiku 4.5
Claude Code users will notice significant speed improvement with Haiku 4.5. Multi-agent projects, rapid prototyping, iterative development—all become more responsive.
Recommended to use Haiku 4.5 for routine tasks (writing tests, code formatting, simple refactoring) and switch to Sonnet 4.5 for complex bugs or architectural decisions.
Claude for Chrome
Claude for Chrome extension works faster and more usefully with Haiku 4.5. Model surpasses Sonnet 4 in computer control tasks while responding more than twice as fast.
For browser task automation—filling forms, data collection, site navigation—Haiku 4.5 provides optimal combination of speed and accuracy.
What Haiku 4.5 Is Best For
Chat assistants and real-time customer support. Low latency and high response quality make interaction natural.
Pair programming and rapid prototyping. Instant answers to questions, code autocomplete, test generation—all work without delays.
Parallel subtask processing as part of the agent team. Sonnet 4.5 plans, several Haiku 4.5 execute work in parallel.
Browser task automation through Claude for Chrome. Best computer control among fast models.
High-load projects where cost is critical. Three times cheaper than Sonnet 4.5 at nearly comparable quality for many tasks.
Conclusion
Claude Haiku 4.5 is the best fast model for programming at the end of 2025 with 73.3% result on SWE-bench Verified. Three times cheaper than Sonnet 4.5 and more than twice as fast, while surpassing Sonnet 4 on some tasks.
Model perfectly suited for real-time tasks, parallel subtask processing and high-load projects. Unique extended thinking support and best security among all Claude models make Haiku 4.5 an excellent choice for a wide range of applications.
For tasks where speed and cost are critical but quality shouldn't suffer, Haiku 4.5 is the optimal solution in the current market.
Frequently Asked Questions (FAQ)
When was Claude Haiku 4.5 released?
Claude Haiku 4.5 was released October 15, 2025 by Anthropic. This is the first Haiku model with extended thinking support and the best fast model for programming at release time.
How much does Claude Haiku 4.5 cost?
Claude Haiku 4.5 price is $1 per million input tokens and $5 per million output tokens. This is three times cheaper than Sonnet 4.5 ($3/$15) and five times cheaper than Opus 4.5 ($5/$25). With caching you can get up to 90% discount.
How does Claude Haiku 4.5 differ from Sonnet 4.5?
Main differences: Haiku 4.5 is three times cheaper ($1/$5 vs $3/$15) and more than twice as fast. Performance is lower (73.3% vs 77.2% on SWE-bench), but on some tasks including computer use, Haiku surpasses even Sonnet 4. Haiku is the best choice for real-time tasks and high loads.
What is Claude Haiku 4.5 result on SWE-bench?
Claude Haiku 4.5 showed 73.3% on SWE-bench Verified—almost at the level of Sonnet 4 (~60%), which was state-of-the-art in spring 2025. Among fast models this is the best result: GPT-5.1 Mini ~65%, Gemini 3 Flash ~55%, GPT-4o mini ~50%.
Does Haiku 4.5 support extended thinking?
Yes, Claude Haiku 4.5 is the first Haiku model with extended thinking support. This is the mode when the model "thinks" before answering, improving quality on complex tasks. Can set a token budget for thinking, for example 128K.
Is Claude Haiku 4.5 free?
Yes, Claude Haiku 4.5 available free to all claude.ai and mobile app users. This is the default model for free tier. For API usage, the standard price of $1/$5 per million tokens applies.
Is Haiku 4.5 safer than other Claude models?
Yes, by automated alignment assessment metric, Haiku 4.5 showed statistically significantly lower levels of problematic behavior than Sonnet 4.5 and Opus 4.1. This is the safest Anthropic model at release time. Classified as ASL-2 (less strict level than ASL-3).
What is the context size of Claude Haiku 4.5?
Claude Haiku 4.5 context window is 200,000 tokens for input and up to 64,000 tokens for output. This is the same size as Sonnet 4.5 (standard configuration) and Opus 4.5.
Is Haiku 4.5 better than GPT-4o mini?
Yes, Haiku 4.5 is significantly more powerful than GPT-4o mini: 73.3% vs ~50% on SWE-bench Verified. However, Haiku is more expensive—$1/$5 vs $0.15/$0.60 for GPT-4o mini. For quality tasks choose Haiku, for a huge volume of simple queries—GPT-4o mini.
Is Claude Haiku 4.5 available through API?
Yes, Claude Haiku 4.5 available through Claude API (identifier claude-haiku-4-5), Amazon Bedrock and Google Vertex AI. The model serves as a direct replacement for Haiku 3.5 and Sonnet 4—simply changing the model identifier in code.
Andrey Shcherbina
Dec 19, 2025
