Claude Opus 4.5: Anthropic attacks Google and OpenAI with price dumping and agent expertise

The AI carousel is spinning faster again and this time the impetus is not coming from Google or OpenAI, but from one of their previously underestimated opponents: Anthropic. On November 24, 2025, the San Francisco-based company officially unveiled its latest language model “Claude Opus 4.5”. And what at first glance might appear to be another “me-too” upgrade turns out to be a targeted offensive on closer inspection. In the sights: GPT-5.1 from OpenAI and Gemini 3 Pro from Google. The goal: technological dominance in the areas of coding, agent control and enterprise integration, flanked by a massive price drop.

Opus 4.5 not only brings more performance, but also more control: a new “Effort Parameter” allows developers to adjust the computing effort depending on the task and desired response quality. This should not only reduce token consumption, but above all costs. And compared to its predecessor, Claude 3 Opus, the costs have literally imploded: instead of 15 US dollars for one million input tokens, Anthropic now only charges 5. The price for output has fallen from 75 to 25 dollars. A declaration of war as it is written in the book, especially to OpenAI, which continues to focus on premium prices with GPT-5.1. But Opus 4.5 doesn’t just want to be cheaper, it wants to be better. In practical benchmarks, such as SWE-Bench Verified, which evaluates real-world coding tasks, the new Claude model beats both GPT-5.1 and Gemini 3 Pro. While Google scores well in visual, multimodal tasks, Anthropic specifically gives its model the ability to shine in real software projects: with robust code, better tool usage (e.g. for Excel, Chrome, local applications) and a systematically improved agent architecture.

Anthropic is thus focusing on a market segment that is only half-heartedly addressed by OpenAI: Automation, software development and enterprise integration. Instead of chasing after every trend – be it text-to-video, AI avatars or social media gimmicks – Opus 4.5 focuses on the backbone of modern knowledge work: clean code, traceable actions and contextual understanding over the long haul. 200.000 token context windows are standard here, not optional. And anyone planning longer conversations or agent workflows will quickly appreciate this extension. Of course, the whole thing remains a double-edged sword. The more autonomously such models can operate, the more urgent the question of security becomes. Prompt injections, i.e. malicious manipulation of user input, are still a real danger. Anthropic talks about improvements in the protection mechanism, but real immunity? Not a chance. The security problem is structural, not cosmetic.

And so the classification remains ambivalent: Claude Opus 4.5 is not an AI firework for consumers, but a sober tool for professionals. A specialist, not an entertainer. If you want to illustrate PowerPoint slides or generate TikTok scripts, you’re better off with GPT or Gemini. However, anyone looking for stable, scalable and controllable results in development environments should take a closer look at Opus 4.5. Especially in an enterprise context, the model could have a disruptive effect in the medium term, not despite, but because of its specialization. With Opus 4.5, Anthropic is positioning itself as the “workhorse” among the major language models. No marketing hysteria, no hype tsunami, but precise differentiation, efficiency and calculable performance. Will that be enough to put pressure on OpenAI and Google? Yes, at least where cost structure, reliability and depth of context count more than visual gimmicks. And it’s a smart move in a market that is increasingly shifting from fascination to function.

If Claude Opus 4.5 proves anything, it is this: The future of AI lies not just in spectacle, but in solid

Source: Anthrop/C