At its first developer conference, Anthropic unveiled two AI models claimed to be the "strongest in the industry," intensifying competition with OpenAI and Google.
Anthropic officially released the Claude4 series of models at its inaugural developer conference on Thursday, including two new models: Claude Opus4 and Claude Sonnet4. The company claims these models lead in multiple popular benchmarks and have been specifically optimized for programming tasks.
Product Features and Pricing Strategy
The newly released Claude4 series has the ability to analyze large datasets, execute long-term tasks, and handle complex operations. Both models excel in programming tasks, making them particularly suitable for coding and editing work.
In terms of user access, Sonnet4 is available to all users, both free and paid, while the more powerful Opus4 is only accessible to paying users. In terms of API pricing, Opus4 costs $15 per million tokens (input) / $75 (output), while Sonnet4 costs $3 (input) / $15 (output). As a reference, one million tokens are approximately equivalent to 750,000 words, exceeding the length of War and Peace.
Performance and Technological Innovation
According to internal tests by Anthropic, Opus4 outperformed Google Gemini2.5Pro and OpenAI's o3 and GPT-4.1 in the SWE-bench Verified programming capability assessment. However, it still fell short of OpenAI's o3 in the multi-modal evaluation MMUU and the PhD-level scientific question test GPQA Diamond.
Both models adopt a "hybrid" architecture that enables quick responses as well as deep reasoning. When the reasoning mode is enabled, the model spends more time thinking about solutions before answering and displays a "user-friendly" summary of the thought process. The models also support parallel use of multiple tools, extracting facts and saving them into "memory" to gradually build an implicit knowledge base.
Development Tool Upgrades
To coincide with the release of the new models, Anthropic significantly upgraded the Claude Code programming tool. The new version is integrated with mainstream IDEs and provides SDK support, allowing developers to connect it to third-party applications. The company also released extensions and connectors for VS Code, JetBrains, and GitHub.
Commercial Ambitions and Market Competition
This launch comes at a time when Anthropic seeks to significantly increase its revenue. It has been reported that this company, founded by former OpenAI researchers, aims to achieve $12 billion in revenue by 2027, a substantial increase compared to an expected $2.2 billion this year. The company recently secured a $2.5 billion credit facility and raised billions of dollars from investors such as Amazon.
Facing challenges from competitors like OpenAI and Google, which continue to release powerful models and development tools, Anthropic has committed to adopting a more frequent model update strategy to maintain a competitive edge.
Security Considerations
It is worth noting that Anthropic has implemented stricter security measures for Opus4, including enhanced harmful content detectors and cybersecurity defenses. Internal testing found that this model could "significantly enhance" the ability of individuals with STEM backgrounds to acquire or deploy dangerous weapons, reaching Anthropic's "ASL-3" safety standard level.
Despite existing security vulnerabilities and errors in AI models regarding code generation, their potential to improve coding efficiency has prompted enterprises and developers to quickly adopt these technologies. As the AI race continues to escalate, whether Anthropic can secure a place for itself in the fierce market competition with the Claude4 series remains worth watching.