Quick Take: Anthropic just launched Claude Opus 4 and Claude Sonnet 4, their next-gen AI models. They’re claiming Opus 4 is the “world’s best coding model,” built for complex, long-haul tasks and AI agent workflows. Sonnet 4 gets a big upgrade too, promising better coding and reasoning. Key new features include “extended thinking” with tool use (like web search), parallel tool execution, and improved memory when given local file access. Plus, Claude Code is now generally available with deeper IDE integrations.
Meet the New Kids: Opus 4 & Sonnet 4
First up, Claude Opus 4. They’re not shy about calling this their most powerful model yet and, get this, the “best coding model in the world.”
It’s designed for sustained performance on those marathon tasks that require thousands of steps and can apparently work continuously for hours. Think complex coding projects and advanced AI agent workflows. Early feedback from places like Cursor, Replit, and Cognition suggests it’s a beast for understanding complex codebases and tackling challenges other models choke on. Rakuten even had it refactor open-source code independently for 7 hours straight.
Then there’s Claude Sonnet 4. This is a significant step up from Sonnet 3.7, delivering better coding chops (hitting a state-of-the-art 72.7% on SWE-bench, just a hair above Opus 4’s score there, interestingly) and sharper reasoning.
It’s positioned as the more balanced option for performance and efficiency, with better steerability. While it might not match Opus 4 across the board, it’s pitched as the practical powerhouse for everyday use cases. GitHub is apparently so impressed they’re making it the engine for the new coding agent in GitHub Copilot.
Both models are hybrids, offering near-instant responses for quick tasks and an “extended thinking” mode for when deeper reasoning is needed.
New Bag of Tricks: Extended Thinking, Parallel Tools & Better Memory
Alongside the new models, Anthropic is rolling out some cool new capabilities.
A big one is extended thinking with tool use (currently in beta). This means both Opus 4 and Sonnet 4 can use external tools, like web search, during their thought process. So, Claude can reason for a bit, grab some info from the web, reason some more, and then use another tool, all to improve its final response.
They’ve also enabled parallel tool execution. This means the models can use multiple tools simultaneously, which should speed things up and allow for more complex interactions.
And here’s a kicker for developers: when you give these models access to local files, they demonstrate significantly improved memory capabilities. They can extract and save key facts over time, building up “tacit knowledge” and maintaining continuity. That’s a big step towards more persistent and context-aware AI.
Claude Code: Now Ready for Prime Time
After a well-received research preview, Claude Code is now generally available.
This means more ways for developers to collaborate with Claude. It now supports background tasks via GitHub Actions and boasts native integrations with VS Code and JetBrains IDEs. Edits from Claude will appear directly inline in your files, making for a smoother pair programming experience. You can install it by running Claude Code in your IDE terminal.
They’re also releasing an extensible Claude Code SDK, so devs can build their own custom agents and apps using the same core agent. As an example, they’ve launched “Claude Code on GitHub” (beta), allowing you to tag Claude Code on pull requests to respond to reviewer feedback, fix CI errors, or modify code.
API Goodies & Availability
For developers building with the Anthropic API, there are four new capabilities: a code execution tool, an MCP connector (interesting!), a Files API (likely tied to that new memory feature), and the ability to cache prompts for up to one hour. These are all aimed at building more powerful AI agents.
The Pro, Max, Team, and Enterprise Claude plans get both Opus 4 and Sonnet 4, including extended thinking. Sonnet 4 is also available to free users, which is a nice touch. Both models are hitting the Anthropic API, Amazon Bedrock, and Google Cloud’s Vertex AI. Pricing is staying consistent with previous versions: Opus 4 at $15/$75 per million input/output tokens, and Sonnet 4 at $3/$15.
It’s a packed announcement, and Claude 4 is clearly aiming to be a major player, especially in the developer and AI agent space.
