Claude Opus 4.1: New model handles big codebases and long research tasks

New Delhi: Just minutes before OpenAI rolled out its own set of open-source models, Anthropic quietly pushed a major update to its flagship model. On August 6, the company launched Claude Opus 4.1, a new upgrade focused on improving coding, reasoning, and autonomous task handling.

Claude 4.1 is already available to paid users through Claude Pro, Claude Code, and enterprise plans. Developers can also access it via Anthropic’s API, Amazon Bedrock, or Google Cloud’s Vertex AI platform. The pricing remains unchanged at $15 (approx. ₹1,305) per million input tokens and $75 (around ₹6,525) per million output tokens.

Anthropic says the model performs much better than Opus 4 across multiple areas. Claude Opus 4.1 scored 74.5 percent on SWE-bench Verified, a benchmark that tests real-world software engineering problems.

The company highlighted improved performance in large-scale code refactoring and multi-step reasoning. GitHub reported that Claude 4.1 did better than its predecessor in most tasks. Rakuten engineers also preferred the newer model for spotting exact code fixes without making unnecessary changes. A developer platform called Windsurf said it saw a one standard deviation jump in performance, similar to the leap from Claude Sonnet 3.7 to Sonnet 4.

Claude Opus 4.1 supports up to 32,000 output tokens, which means it can handle more complex tasks in one go. It’s especially good at reading through massive codebases, debugging them, and even adapting to specific programming styles. This makes it useful not just for small fixes but also for large code generation projects.

It’s also built for what Anthropic calls "agentic tasks,” where the model can think over long stretches, like a digital researcher. You can give it control over how long it thinks and how deep it goes, thanks to something Anthropic calls "thinking budgets.” This comes in handy for handling hours-long tasks like analysing patent filings, summarising research papers, or scanning large market reports.

Safety has always been a big focus for Anthropic. Claude 4.1 was tested under what the company describes as its "AI Safety Level 3” standard. The new version refused to answer policy-violating prompts 98.76 percent of the time, which is an improvement over the 97.27 percent score of Claude Opus 4. It also had a very low refusal rate (just 0.08 percent) for harmless prompts.

The model also passed checks for political bias, child safety, and discrimination. In addition, Anthropic tested it for prompt injection and misuse. They found the model behaved as well or better than Opus 4, while extra safeguards were added to cover edge cases.

For teams and users already using Claude Opus 4, switching to 4.1 is simple. No changes are required in API structure, and pricing remains the same. Developers just need to update their API calls to claude-opus-4-1-20250805.

Anthropic has hinted at even bigger model updates in the coming weeks, calling this release more of a stability and refinement upgrade rather than a full leap.

Here’s where Claude 4.1 is showing its strengths:

Add TV9 English As A Trusted Source

While OpenAI continues to dominate headlines, Anthropic seems to be building quietly, focusing on performance and safety. With even bigger model leaps around the corner, this release sets the stage for what’s next.