Anthropic unveils Claude 4 models to take on OpenAI and Google in the Agentic AI race

AI model launches are arriving faster than ever. OpenAI recently debuted GPT-4.1 and its new o3 reasoning models, Google followed with Gemini 2.5 Pro, and now Anthropic has entered the fray with Claude 4 — its most advanced family of models designed not just to respond, but to reason, code, and carry out sustained, complex work.

Unveiled at its Anthropic’s Code with Claude developer event, Anthropic introduced two models: Claude 4 Opus and Claude 4 Sonnet, both part of what the company calls a major leap forward in intelligence, usefulness, and autonomy.

According to Anthropic, these models “set a new standard for AI agents,” with capabilities that include analysing thousands of documents, executing long-horizon tasks, and are tuned to perform well on programming tasks, making them well-suited for writing and editing code.

Claude Opus 4, the flagship, is positioned as the best coding model in the world, according to Anthropic. It scored 72.5% on SWE-bench, a benchmark for real-world software engineering tasks, beating Google’s Gemini 2.5 Pro and OpenAI’s o3 and GPT-4.1. It also scored 43.2% on Terminal-bench, which evaluates command-line performance. Anthropic says Opus can autonomously work for up to seven hours, maintaining context and producing consistent output across multi-step challenges.

Claude Sonnet 4, while more lightweight, still delivers strong performance. It builds on Claude 3.7 – launched earlier this year– with better control and instruction following. CEO Dario Amodei noted that the earlier Sonnet often overdid tasks — an issue addressed in version 4. Sonnet 4 now balances capability and efficiency and performs comparably to Opus on some coding benchmarks.

But beyond the raw performance gains, both Claude 4 models arrive with expanded autonomy. Both models introduce tool use and improved memory, enabling Claude to retrieve information on the web mid-task, access files, and retain key details across sessions. These capabilities, still in beta, support Anthropic’s push beyond reasoning toward what Anthropic describes as “agentic” AI — models that don’t just respond but actively plan, reason, and act.

Anthropic also rolled out updates to Claude Code, its terminal-native coding assistant, allowing it to integrate with IDEs like VS Code and JetBrains, and supports GitHub Actions for background tasks and file editing.

With the launch, Claude 4 is now accessible via Claude.ai, Amazon Bedrock, and Google Cloud’s Vertex AI. Sonnet 4 is available to all users, while Opus 4 is reserved for paid plans. API pricing starts at $15 per million input tokens and $75 for output for Opus, while Sonnet costs $3 and $15, respectively.

As competition intensifies, Anthropic is positioning itself not just as another model vendor but as a platform for real AI work. It shows in its financial performance. Anthropic’s revenue doubled to a $2 billion annual run rate in Q1. With a recent $2.5 billion credit line, the company is signalling it’s ready to compete for the long haul, not just with smarter models, but more useful ones.