Claude 4 models
Claude Opus 4 and Claude Sonnet 4 are the newest generation of AI models from Anthropic, an Amazon-backed AI safety and research firm created by former OpenAI research leaders. These models, which were unveiled somewhere between May 22 and 23, 2025, are said to be establishing “new standards for coding, advanced reasoning, and AI agents“. Anthropic reportedly stopped investing in chatbots at the end of the previous year to concentrate on enhancing Claude’s capacity to handle complex tasks like research and coding, so the launch marks a daring move away from competing only in the chatbot space and towards becoming a well-known AI coding platform.
Claude Opus 4
It is stated that Claude Opus 4 is the “best coding model in the world” and that it is Anthropic’s most potent model to date. It can operate independently for lengthy periods of time, including almost a complete corporate workday (seven hours) in customer testing, and it exhibits consistent performance on intricate, time-consuming tasks and agent workflows. It is advised to utilize Opus 4 for difficult use cases that call for “frontier intelligence,” like:
- Sophisticated AI agents.
- Complex coding tasks, such as creating full-stack apps and reworking big codebases.
- Agentic search, research synthesis, and deep research problems.
- Independent work with a long time horizon that prioritizes competence and precision.
- Development of content with an emphasis on natural writing and of human quality.
Claude Sonnet 4
Anthropic’s mid-size model, the Claude Sonnet 4, strikes a balance between price and performance. It is a major improvement to Claude Sonnet 3.7, which it replaced. It offers “superior coding and reasoning,” more accurate replies, and a 65% lower likelihood of “reward hacking.” High-volume and broad activities are appropriate for Claude Sonnet 4 , such as:
- Coding activities such as bug fixes and code reviews.
- AI helpers for in-the-moment client communications.
- Effective research and analysis, such condensing market signals or dashboards.
- Extensive production and analysis of material.
- Functioning in multi-agent systems as a task-specific subagent.
Both Claude Sonnet 4 and Opus 4 are hybrid reasoning models that may provide answers almost instantly while also allowing for deeper reasoning through the use of a “extended thinking” mode. The models perform better on challenging tasks with this longer thinking mode, which gives them more time to think through potential solutions. The models can display a “user-friendly” synopsis of their reasoning process. A new Developer Mode gives you access to whole, unprocessed sequences of thinking.
Both models now have the ability to use parallel tools (such as online search), which enables them to contact several APIs or plugins at once to expedite processes and lower errors. If given access to local files, they can also extract and store important information to create “memory files” or “tacit knowledge” over time, increasing continuity and dependability on long-term activities. Additionally, the models are more accurate at following directions.
Opus 4 and Claude Sonnet 4 have achieved industry-leading results on the SWE-bench coding benchmark, demonstrating the significant emphasis on coding. Additionally, Opus 4 excels at real-world coding tasks on the Terminal Bench test. It is pointed out that performance varies and that internal benchmarks should be regarded “with a grain of salt.” In specialized activities like high school arithmetic, certain benchmarks display regressions when compared to previous models. Even while AI models currently have trouble writing high-quality software and sometimes introduce flaws or errors, their potential to increase productivity is propelling their quick adoption.
Anthropic has made its Claude Code agentic command-line tool widely accessible to assist developers. Edits may be seen immediately in files with Claude Code’s integration with well-known IDEs like GitHub, Microsoft’s VS Code, and JetBrains. There is also an expandable Claude Code SDK for creating unique agents and apps. A code execution tool, an MCP connection, the Files API, and prompt caching are examples of new API features.
Opus 4 and Claude Sonnet 4 may be accessed using Google Cloud’s Vertex AI platform, Amazon Bedrock, and the Anthropic API. Databricks customers can also access them natively. Opus 4 is part of the premium Claude plans (Pro, Max, Team, and Enterprise) on Anthropic’s own platform, whereas Claude Sonnet 4 is accessible to both free and paying users. Claude Sonnet 4 has a beginning price of $3 per million input tokens and $15 per million output tokens, whereas Opus 4 has a starting price of $15 per million input tokens and $75 per million output tokens. Batch processing combined with timely caching can save costs.
To make sure the models satisfy safety, security, and dependability requirements, Anthropic has carried out a thorough testing and assessment process in collaboration with other specialists. They are made available with more stringent security measures, such as strengthened cybersecurity and dangerous content detection systems. Although internal testing indicated that Opus 4 may “substantially increase” the capacity of someone with a STEM background to access, develop, or deploy chemical, biological, or nuclear weapons, the models are evaluated against Anthropic’s “ASL-3” model definition.
Positive early client feedback has been received from businesses such as Palo Alto Networks, Replit, Cursor, Rakuten, Augment Code, and others, who have reported increases in agent performance, complicated job management, coding velocity, and code quality. Claude on Vertex AI, for instance, increased code development pace at Palo Alto Networks by 20% to 30%. Block pointed out that Opus 4 was the first model to improve code quality while debugging and editing without compromising speed.
In terms of finances, Anthropic said that the first quarter of 2025 saw $2 billion in annualized sales, more than double the previous quarter. In 2027, the corporation wants to make $12 billion. Wall Street is still making investments; in anticipation of growing development expenses, Anthropic has raised billions from investors like as Amazon and has a $2.5 billion credit line.
In order to stay competitive and make enhancements more quickly, Anthropic also intends to switch to more regular model upgrades.
