Claude Opus 4.6 Release Update: Anthropic has just dropped its latest artificial intelligence model on Thursday, called the Claude Opus 4.6, and this brings an update to its foundational Opus 4 model. This new upgraded version brings in some enhancements over its base model and works on limitations like sustained and improved performance when it comes to complex tasks, especially when software engineering and knowledge-laden workloads are considered. Claude 4.5 is quite capable of handling complex tasks, but it does have its own limitations, often faltering with edge cases and long-horizon contexts needed to execute long-term planning and execution. A beta version of Opus 4.6 has fixed that by increasing the context window to one million tokens, which roughly translates to 750,000 words, which basically means bigger and larger databases can now be analysed.

ALSO READ: iPhone 18 Pro And iPhone 18 Pro Max Could Focus More On Performance Than Design: Reports Suggest

What Is The New Model All About?

In its newsroom post, Anthropic has laid down the details of its latest model. Anthropic says it's the first time their Opus model will come with one million tokens for its context window; this means that users can now input considerably larger amounts of data without hindering long-duration interactions with better precision. This is a significant upgrade from the previously available 200,000 tokens that Opus models had earlier.

This new model also comes with its array of new features, like the context compaction that can aid in summarising and refreshing data during long and complex tasks. The model also comes with adaptive thinking that delivers a much more nuanced approach when it comes to tasks requiring deeper reasoning if needed and effort controls from low to max to optimise cost, speed and reasoning intelligence.

What The Benchmarks Say

On the benchmarking front, Claude Opus 4.6 is clearly pushing the bar higher. Based on the company’s own internal tests, the model comes out on top in Terminal-Bench 2.0 for command-line skills and Humanity’s Last Exam for broad, cross-domain reasoning.

It also pulls ahead in agent-style evaluations such as GDPval-AA, where it reportedly beats OpenAI’s GPT-5.2 by roughly 144 Elo points and improves on Opus 4.5 by about 190 points, especially in finance and legal tasks. SWE-bench Verified scores average 81.42 per cent with optimised prompting, while CyberGym results highlight strong baseline performance even without explicit reasoning steps.

As far as safety is concerned, Anthropic states that this model has a higher rating than its peers in safety audits, with reduced sycophancy and deception, and also lower rates of refusal tendencies as compared to recent releases. The company has said that six new cybersecurity probes have been added to find out any potential misuse.

As far as coding is concerned, the latest Opus 4.6 model is able to manage large repositories autonomously; apart from that, it can also effectively conduct code reviews and can perform highly accurate debugging. It can also spin up agent teams for parallel development through the research preview of Claude Code. On the business side, it handles financial analysis, document creation, and multi-step research using tools like Claude in Excel, which now supports unstructured data and longer workflows. A research preview of Claude in PowerPoint brings the same capabilities to presentations. In more specialised areas such as computational biology, it reportedly delivers nearly twice the performance of Opus 4.5, helping speed up scientific research.

ALSO READ: Planet Parade 2026: This February Watch Out As Celestial Bodies Align, When, Where And How To Watch

Opus 4.6 Pricing And Availability

This latest version can be found on Anthropic's official website, as well as mobile and desktop apps, Anthropic's application programming interface (API) and major cloud providers. The pricing at which the API starts is $5, which roughly translates to Rs 453 per million input tokens, and $25 (approx.Rs. 2,300) for output, with premium tiers offering an extended context window.