Claude Opus 4.6: Discover What’s New and Impressive

Introducing Claude Opus 4.6. ELO scores better than Opus 4.5. It's clearly better than the previous models, even Gemini 3 Pro and GPT 5.2. Agentic terminal coding, agentic computer use, agentic tool use, agentic search, and for most of these it's topping the list.

Long context retrieval that is the best software failure diagnosis. Opus 4.6 is the best in regards to accuracy misalignment behavior.

Claude Opus 4.6: Discover What’s New and Impressive Benchmarks

• ELO scores better than Opus 4.5.
• Better than the previous models, even Gemini 3 Pro and GPT 5.2.
• Topping the list for agentic terminal coding, agentic computer use, agentic tool use, and agentic search.
• Strong long context retrieval and the best software failure diagnosis.
• Best in regards to accuracy misalignment behavior.

Claude Opus 4.6: Discover What’s New and Impressive Key Features

• 1 million token context window.
• Improved abilities in everyday work tasks such as financial analysis, doing research, creating documents, and spreadsheets.
• Claude in Excel and PowerPoint demo presentations.
• Multitask autonomously in Claude Code.
• Assemble agent teams or multiple agents working together to complete a complex task.

Claude Opus 4.6: Discover What’s New and Impressive in Excel and PowerPoint

You can see a major upgrade for Claude in Excel. Similarly, Claude in PowerPoint can automate creating PowerPoint presentations and creating charts and graphs. Just give the numbers and Claude can automatically build all those things for you. Do not spend time in creating documents and presentations when you have Claude and this is powerful.

Step-by-step - Automate documents and slides with Claude:

1. Install Claude in Excel.
2. Install Claude in PowerPoint.
3. Provide the numbers or content you want to include.
4. Ask Claude to create the presentation, charts, and graphs.

Claude Opus 4.6: Discover What’s New and Impressive Pricing and Availability

You can start using Claude Opus 4.6 using the Claude API.

• Input tokens: $5 per million tokens
• Output tokens: $25 per million tokens

Claude Opus 4.6: Discover What’s New and Impressive Effort Parameter

Currently you have an effort parameter you can control. If you want it to think more then you can increase that. If you think the model is thinking more and you want a quick response then you can decrease the effort.

Step-by-step - Control thinking depth with Effort:

1. Set Effort to a higher level for deeper thinking on complex tasks.
2. Set Effort to a lower level for quicker responses on simple tasks.
3. Choose from four levels based on the amount of effort needed to complete the task.

Claude Opus 4.6: Discover What’s New and Impressive on Context and Memory

From GitHub and developer reports, Opus 4.6 is delivering on complex multi-step coding tasks. There’s a common complaint called context rot where performance degrades as conversations exceed a certain number of tokens. When you keep on talking to your models or any AI agents, you can see a degradation in the quality of the response and it could even forget the initial conversation. In regards to Opus 4.6, it shows better performance compared to the previous models and it is better in the needle in the haystack test.

We also have context compaction, which means compressing the whole context without missing key important points.

• Input tokens: 1,000,000
• Output tokens: 128,000

Claude Opus 4.6: Discover What’s New and Impressive Agent Teams

They are introducing agent teams. When multiple agents start working together to complete a complex task, the same approach is here, orchestrating teams of Claude Code sessions.

Claude Opus 4.6: Discover What’s New and Impressive Live Examples

Claude Code Drawing

You can see Claude Code automatically drawing and it's really impressive to see an AI model trying to draw and explain things. Next time if you want to draw any diagram you don't need to do anything manually, just ask this AI and it should be able to draw things for you in a more interactive way.

Step-by-step - Ask Claude to draw a diagram:

1. Describe the diagram you want.
2. Ask Claude Code to draw and explain the elements.
3. Review the generated diagram and request adjustments if needed.

Computer Use: Automated Image Recreation

It can automatically go and view an image. It saw an image, then it tried to draw the image. Opening Paint and choosing the colors, clicking icons and then getting the background, then slowly drawing and adding elements like boats and sun. The final output looks really nice, an AI agent automatically drawing it for you.

Step-by-step - How the automated recreation proceeds:

1. Provide or reference the image to recreate.
2. Let the agent open Paint.
3. It chooses colors and sets the background.
4. It draws the shapes and layout.
5. It adds elements like boats and sun.
6. Review the final output.

UI/UX Design Capabilities

The whole layout is being built by Claude Opus 4.6 and this looks really nice. The quality of the UI is really high.

Claude Opus 4.6: Discover What’s New and Impressive Integration Plans and Testing

I’m really impressed and I’m definitely going to try with various model providers, integrating that within my own application and also trying incursor, VS Code, wind surf and gravity and much more.

Final Thoughts

Claude Opus 4.6 is better than the previous models on key benchmarks, with strong agentic abilities, a 1 million token context window, adaptive thinking with an effort parameter, and context compaction. It automates documents and presentations in Excel and PowerPoint, draws diagrams through Claude Code, recreates images through computer use, and builds high-quality UI layouts. The pricing is straightforward at $5 per million input tokens and $25 per million output tokens.