Claude 4 Explained: Opus 4, Sonnet 4 and Tool Use
Anthropic launched Claude Opus 4 and Claude Sonnet 4 in May 2025 as hybrid reasoning models, adding tool use during extended thinking and new developer features for agent workflows.
Timeline
- May 22, 2025: Anthropic introduced Claude Opus 4 and Claude Sonnet 4.
- At launch: Claude Code became generally available and Anthropic released new API tools for agent applications.
- After launch: Newer Claude models arrived, making the Claude 4 announcement a historical product snapshot.
Anthropic introduced Claude Opus 4 and Claude Sonnet 4 on May 22, 2025. It positioned Opus 4 as the more capable option for demanding coding and long-running agent tasks, while Sonnet 4 was a lower-cost successor to Claude 3.7 Sonnet. Both were hybrid models that could answer quickly or use an extended-thinking mode for harder problems. [1][2]
The most visible workflow change was tool use during extended thinking. In beta, the models could alternate between reasoning and tools such as web search before producing a final answer. Anthropic also said the models could execute independent tools in parallel, allowing an application to gather several pieces of information without waiting for each request to finish in sequence. [1]
Anthropic described better memory behavior when developers gave a model access to local files. The model could extract facts and save notes that a surrounding application later supplied again. This was not permanent personal memory built into every Claude conversation; it depended on an application providing storage, permissions and the saved files as part of the working environment. [1]
The launch also expanded Anthropic's developer platform. Claude Code moved from research preview to general availability, with integrations for common code editors and support for background work through GitHub Actions. New API capabilities included code execution, an MCP connector, a Files API and longer prompt caching, all aimed at applications that perform multi-step work rather than one isolated response. [1]
At launch, Anthropic listed Opus 4 and Sonnet 4 on its API, Amazon Bedrock and Google Cloud Vertex AI. Sonnet 4 was also available to free Claude users, while paid plans included both models and extended thinking. The launch prices were stated per million input and output tokens, with Opus priced above Sonnet because it targeted more demanding workloads. [1]
Anthropic published coding, reasoning and agent benchmarks and described Opus 4 as its strongest coding model at the time. The company explained that some reported scores used extended thinking while others did not. These were vendor-reported evaluations under particular prompts and tools, so they should be treated as evidence about tested conditions rather than a guarantee for every repository or task. [1][2]
Claude 4's launch mattered because it joined reasoning and action in one product family: a model could spend more compute planning, call tools while doing so and maintain working notes through application-managed files. Those features still required careful permissions, validation and human review. Current buyers should consult Anthropic's live model documentation because model names, availability and pricing have changed since May 2025. [1][2]