Anthropic Raises the Bar with Opus 4.5: A New Standard in AI Performance and Affordability
- Anthropic has launched Opus 4.5, a flagship model that outperforms competitors like GPT-5.1 and Gemini 3 on major coding benchmarks while cutting pricing by roughly 66% to increase market accessibility..
- The model introduces "infinite chat" capabilities and self-improving agentic behaviors, allowing it to handle complex, long-term workflows by intelligently managing memory and refining its own strategies in real-time.
- New seamless integrations with Chrome and Excel, combined with token-saving efficiency updates, position Opus 4.5 as a highly practical tool for enterprise automation and professional software engineering.
Anthropic has officially concluded its 4.5 model series rollout with the launch of Opus 4.5, a flagship artificial intelligence model that promises to redefine the competitive landscape. Released on Monday, this new model arrives shortly after the debuts of Sonnet 4.5 and Haiku 4.5, completing a rapid-fire release schedule designed to counter moves by rivals like OpenAI and Google. Beyond just raw power, Anthropic has made a strategic pivot by slashing prices significantly, making this high-end frontier model far more accessible to developers and enterprises than its predecessor.
In terms of technical capability, Opus 4.5 has demonstrated exceptional results, claiming the top spot on several respected leaderboards. Most notably, it is the first model to surpass the 80% threshold on SWE-bench Verified, a rigorous test for software engineering tasks. This places it ahead of formidable competitors such as OpenAI’s GPT-5.1 and Google’s Gemini 3. Even more strikingly, on Anthropic’s own internal assessments, Opus 4.5 outperformed every human candidate who has ever taken their engineering performance exam, signaling a major leap in AI reasoning and coding proficiency.
To complement this raw intelligence, Anthropic has introduced significant improvements to how the model handles memory and context. Users can now access an "infinite chat" feature, which solves the frustration of running out of space during long conversations. Instead of stopping when the context window is full, the model intelligently compresses past information, allowing it to retain key details over extended sessions. This is particularly vital for "agentic" workflows where the AI acts as a lead manager, directing smaller sub-agents to complete complex, multi-step tasks without losing track of the overall objective.
The release also focuses heavily on practical integration into daily workflows. Anthropic is expanding the availability of Claude for Chrome and Claude for Excel, moving the AI out of the chatbox and directly into the tools professionals use most. The Excel integration, now available to Max, Team, and Enterprise users, supports advanced features like pivot tables and file uploads. Meanwhile, the browser extension allows the AI to assist with web-based tasks more fluidly, further bridging the gap between an intelligent chatbot and a functional work assistant.
Perhaps the most aggressive move in this announcement is the pricing strategy. Anthropic has reduced the cost of Opus 4.5 by approximately two-thirds compared to the previous version, pricing it at $5 per million input tokens. This efficiency is driven by architectural improvements that allow the model to use up to 76% fewer tokens to achieve results comparable to older models. By introducing an "effort parameter," developers can now balance cost against performance, choosing how much computational power the model should dedicate to a specific problem.
Early adopters in the enterprise sector are already reporting transformative results, particularly regarding "self-improving agents." Companies like Rakuten have observed that Opus 4.5 can iteratively refine its own strategies for completing tasks, effectively learning from experience within a session. Unlike traditional learning where the model's core data changes, this involves the AI optimizing its approach in real-time to solve problems faster and more accurately. This ability to "learn on the fly" is proving invaluable for complex coding and administrative automation.
The launch of Opus 4.5 intensifies the ongoing "AI arms race" hitting the tech industry in late 2025. With OpenAI recently releasing specialized coding models and Google updating its Gemini line, the market is becoming increasingly crowded. However, by combining state-of-the-art engineering performance with a dramatic price cut, Anthropic is positioning itself not just as a research leader, but as the most practical choice for businesses looking to deploy AI at scale.
As these models begin to match or exceed human performance on specific technical exams, the conversation is shifting from simple productivity tools to the future of professional work. While Anthropic executives note that human qualities like collaboration are still unique to people, the sheer technical dominance of Opus 4.5 serves as a strong signal that the nature of white-collar jobs, particularly in software engineering, is on the verge of a significant transformation.