Anthropic introduced its latest artificial intelligence model, Claude Opus 4.5, on Monday, marking a significant milestone in AI development with notable improvements in performance and cost efficiency. The company, backed by Amazon, has reduced pricing by approximately two-thirds compared to its previous model, making advanced AI capabilities more accessible to developers and enterprises alike.
The new model demonstrated exceptional proficiency by scoring higher than any human candidate in Anthropic’s most demanding internal engineering exam. This achievement highlights the rapid evolution of AI systems and raises important considerations about the future impact of AI on professional white-collar jobs.
Claude Opus 4.5: Cutting-edge Performance and Competitive Pricing
Anthropic priced Claude Opus 4.5 at $5 per million input tokens and $25 per million output tokens, a steep decline from the $15 and $75 rates of its predecessor, Claude Opus 4.1. This pricing strategy aims to broaden the model’s reach while intensifying competition with major players like OpenAI and Google.
Alex Albert, Anthropic’s head of developer relations, emphasized the company’s focus on practical utility: “We want to make sure this really works for people who want to work with these models. Our goal is to enable Claude to assist users in accomplishing tasks they might prefer not to handle themselves.”
Outperforming Rivals and Human Experts
Internal testing revealed that Claude Opus 4.5 achieved 80.9% accuracy on SWE-bench Verified, a benchmark for real-world software engineering tasks. This result surpasses OpenAI’s GPT-5.1-Codex-Max (77.9%), Anthropic’s Sonnet 4.5 (77.2%), and Google’s Gemini 3 Pro (76.2%). Beyond benchmarks, testers reported that the model exhibits enhanced judgment and intuition, effectively understanding priorities and producing coherent summaries aligned with user needs.
In a notable demonstration, the model outscored all human candidates on Anthropic’s toughest engineering test, designed to assess technical skills and decision-making under time constraints. Using parallel test-time compute, Claude Opus 4.5 achieved the highest recorded score, even matching the best human performance without time limits in Anthropic’s coding environment.
Efficiency Gains and Customizable Performance
Claude Opus 4.5 also offers dramatic efficiency improvements, using up to 76% fewer tokens than previous models to deliver equal or better results. At medium effort, it matches the prior model’s top scores while significantly reducing token consumption; at maximum effort, it outperforms earlier versions while maintaining notable efficiency.
Anthropic introduced an “effort parameter” allowing developers to balance computational intensity with latency and cost, giving users greater control over AI performance.
Early enterprise users have validated these efficiency claims. Michele Catasta, president of Replit, noted that Opus 4.5 solves problems more efficiently than competitors, which compounds benefits at scale. Similarly, GitHub’s chief product officer, Mario Rodriguez, highlighted the model’s suitability for complex coding tasks like migration and refactoring, noting substantial token savings.
Self-Improving AI Agents and Expanded Capabilities
One of the most innovative features of Claude Opus 4.5 is its ability to act as a self-improving agent. Early adopters like Rakuten reported that AI agents powered by Opus 4.5 autonomously refined their capabilities through iterative learning, achieving peak performance in fewer iterations than other models.
Albert clarified that these agents do not alter the model’s fundamental parameters but optimize their problem-solving strategies over time. This capability extends beyond programming to generating professional documents, spreadsheets, and presentations, marking the largest generational leap Anthropic has observed.
Financial firm Fundamental Research Labs also reported improvements, citing a 20% accuracy increase and 15% efficiency gain on internal evaluations, enabling the handling of more complex tasks.
New Features Enhance User Experience and Developer Tools
Alongside the model launch, Anthropic expanded its product offerings for enterprise users. Claude for Excel now supports pivot tables, charts, and file uploads, while the Chrome extension is available to all Max tier users.
Significantly, Anthropic introduced “infinite chats,” which removes context window limits by summarizing earlier conversation content, effectively providing an unlimited context memory. For developers, new “programmatic tool calling” allows Claude to write and execute code that directly invokes functions, and Claude Code’s updated “Plan Mode” enables parallel AI agent sessions on desktop.
Intensifying Competition in AI Market
Anthropic’s release of Claude Opus 4.5 comes amid fierce competition as OpenAI and Google continue rapid innovation. OpenAI has rolled out multiple GPT-5 variants, including Codex Max, capable of autonomous operation for up to 24 hours, while Google recently launched Gemini 3. Anthropic attributes much of its accelerated development to leveraging Claude itself to enhance product and research workflows.
Despite aggressive pricing potentially squeezing margins, Anthropic anticipates broader market adoption, expecting startups to integrate the model extensively into their products.
While profitability remains challenging due to substantial infrastructure and talent investments, the AI market is expected to exceed $1 trillion in revenue within a decade. No single company yet dominates, even as AI models begin to automate complex knowledge work.
Industry leaders have praised Opus 4.5’s advancements. Michael Truell, CEO of Cursor, described it as a notable improvement in pricing and intelligence for difficult coding tasks, while Scott Wu, CEO of AI coding startup Cognition, highlighted its reliable performance during extended autonomous coding sessions.
Implications for the Future of Work
As AI models like Claude Opus 4.5 approach or surpass human expert performance on technical tasks, their influence on professional industries becomes more immediate and tangible. Alex Albert emphasized that these developments are significant signals for how AI will integrate into workplace roles, particularly in engineering, and warrant close attention.
“I think it’s a really important signal to pay attention to,” Albert concluded, underscoring the transformative potential of these AI advancements.
Fonte: ver artigo original
Related Articles
Anthropic Unveils Cowork: A User-Friendly AI Desktop Agent for File Management Without Coding
European Commission Launches Antitrust Investigation into Meta’s WhatsApp AI Policy
IBM Highlights Data Silos as Major Obstacle to Scaling Enterprise AI- Korean AI Startup Motif Shares Four Key Lessons for Training Enterprise Large Language Models
