Anthropic launches Claude Sonnet 4.5, the top AI for programming
Anthropic has just unveiled the latest iteration of its flagship language model, Claude Sonnet 4.5. This version claims to be the best artificial intelligence for programming in the world, setting a new benchmark by surpassing competitors like OpenAI's GPT-5 and Google's Gemini 2.5 Pro.
Beyond programming, Claude Sonnet 4.5 boasts significant enhancements in mathematics and reasoning capabilities. It also excels at controlling computers and executing automated actions, positioning it as a powerful tool for complex AI agents.
Performance evaluations in software engineering tasks place Claude Sonnet 4.5 ahead of its rivals and its predecessors. This new model has achieved a precision rating that is higher than that of Claude Opus 4.1, previously regarded by Anthropic as their most capable AI for programming.
In tests conducted on SWE-bench Verified, Claude Sonnet 4.5 attained a precision score of 77.2% (82% when utilizing parallel time testing), compared to 74.5% for Opus 4.1, 72.8% for GPT-5, and 67.2% for Gemini 2.5 Pro.
Moreover, Anthropic emphasizes that Claude Sonnet 4.5 demonstrates improved reasoning skills and optimized performance when dealing with specialized knowledge across various fields such as finance, law, and medicine, to name a few.
Claude Sonnet 4.5: the most advanced AI by Anthropic to date
A notable aspect of Claude Sonnet 4.5 is that its developers describe it as the "most aligned" AI they have created. This term refers to its lower propensity for engaging in undesirable behaviors such as deception, promoting delusions, following harmful instructions, or seeking power.
What stands out is that this new AI from Anthropic not only performs better than its rivals from OpenAI and Google but also shows considerable improvements over its predecessors. Claude Sonnet 4.5 is significantly less likely to exhibit unwanted behaviors compared to Opus 4.1 and Sonnet 4, marking a substantial advancement in just a few months.
As expected, Claude Sonnet 4.5 is accessible not only through the Claude chatbot but also via an API. The good news for developers is that pricing remains unchanged from Sonnet 4, set at $3 for every million tokens of input and $15 for each million tokens of output.
Additionally, Anthropic has introduced a new SDK called Claude Agent SDK, designed to simplify the creation of AI agents. This tool leverages the same technology as Claude Code but extends beyond tasks related to software engineering.
Lastly, the team behind Claude Sonnet 4.5 has also released the highly anticipated Claude for Chrome browser extension. However, it is essential to note that this utility is currently available only to those who pay for Claude Max (starting at $100 per month) and have signed up for the waitlist.
Key features of Claude Sonnet 4.5
- Advanced programming capabilities: It outperforms leading models in software development tasks.
- Improved mathematical reasoning: Enhanced skills in complex mathematical calculations.
- Specialized knowledge handling: Better performance in fields such as finance, law, and medicine.
- Alignment and safety: Decreased likelihood of harmful behaviors compared to previous models.
- Accessibility: Available through both a chatbot interface and API.
- Cost-effective for developers: Competitive pricing structure remains the same as prior versions.
- New development tools: Introduction of Claude Agent SDK for easier AI agent creation.
Performance comparisons with competitors
When comparing Claude Sonnet 4.5 with its main competitors, it is evident that it has set new standards in several areas. The performance metrics not only reflect superior precision but also highlight the efficiency of the model in handling various tasks.
| Model | Precision Score |
|---|---|
| Claude Sonnet 4.5 | 77.2% (82% parallel testing) |
| Claude Opus 4.1 | 74.5% |
| GPT-5 | 72.8% |
| Gemini 2.5 Pro | 67.2% |
Implications for developers and businesses
The launch of Claude Sonnet 4.5 offers significant implications for developers and businesses alike. With advanced programming capabilities and improved alignment features, organizations can leverage this AI for various applications, enhancing productivity and reducing risks associated with AI behavior.
For developers, the continued availability of the API at the same cost as previous models enables them to integrate this advanced technology into their applications without incurring additional expenses. This is a considerable advantage, especially for startups and small businesses that rely heavily on cost-effective solutions.
Moreover, the introduction of the Claude Agent SDK represents a valuable resource for developers looking to create sophisticated AI agents without the steep learning curve typically associated with such tasks. This could lead to a rise in innovative applications across different sectors.
For those interested in a deeper comparison of AI models, this informative video discusses the differences and performances of Claude and its competitors:
Leave a Reply