The AI arena is buzzing again, and this time, the news is all about Anthropic’s latest power move: the highly anticipated debut of Claude 4! After a week of whispers surrounding a system prompt leak, this launch is a clear statement, grabbing the industry’s attention with both hands.
The stars of the show are two promising new models: Claude Opus 4 and Claude Sonnet 4. Upon release, they immediately raised the bar for programming, reasoning, and AI agents, solidifying their place as industry leaders.
Seven-Hour Code Marathon with No Breaks
The most talked-about model in this upgrade is undoubtedly the flagship, Claude Opus 4. Its appeal goes beyond mere “intelligence”; it also showcases incredible endurance and a relentless drive to achieve, a characteristic that’s setting it apart.
Rakuten, a global leader in enterprise software, tasked Opus 4 with a complex and rigorous open-source code refactoring project. Under intense scrutiny, Opus 4 demonstrated remarkable autonomy, churning out code continuously for a staggering seven hours. Its performance remained remarkably stable throughout, showing no signs of slowing down.
This single example alone is enough to impress the industry, highlighting its long-duration capabilities and its impressive grasp of complex contextual information.
This “extended battery life,” combined with deep memory and precise planning, means that it can excel at extremely complicated tasks that require long-term commitment and multi-step reasoning, understanding and executing those intricate plans step-by-step.
For example, in the advanced challenges of the now-famous “Claude plays Pokémon” experiment, Opus 4 exhibited superior long-term memory and strategic planning compared to its predecessors, capable of playing for 24 hours straight, in stark contrast to the previous 45-minute limit of previous models.
Claude 4’s New Tech: Hits That Matter
This upgrade to Claude 4 isn’t just about numbers; it’s packed with notable technological advancements, each potentially redefining AI applications. Firstly, its core “dual-mode” thinking mechanism turns Opus 4 and Sonnet 4 into versatile, hybrid performers.
They can provide immediate, near-instant responses to your commands, perfect for those who prioritize speed. Meanwhile, when faced with intricate, complex problems, they can smoothly switch to “extended-thinking” mode, dedicating higher computational power for longer and more in-depth reasoning to arrive at the most thorough and accurate solutions.
Moreover, Opus 4’s new “external” memory feature, charmingly named “memory files,” is a significant highlight. With the right permissions, it can intelligently extract key information from local files when processed and safely store them in these dedicated “memory files.”
This means it no longer loses track during lengthy conversations or projects lasting days, a revolutionary benefit for complex applications that require consistent monitoring and context.
Furthermore, the new models now have strong tool usage capabilities (currently in beta testing), which means Claude 4 can “invoke tools” when needed.
When tackling complex problems and feeling like it lacks sufficient knowledge, it can do what human experts do: actively call up external tools like web search to gather the latest information or specific data, allowing for real-time learning. It can even coordinate the use of several tools at once to work together, substantially expanding its problem-solving abilities.
Of course, improved user experience also depends on maximizing the command interpretation capabilities. The new models now understand your intentions better, and complex commands that might have previously stumped AI can now be accurately understood.
They have also developed a helpful “thought summarization” skill: after any extremely complex thinking processes, the system will use a relatively compact model to condense lengthy, convoluted thought processes into a concise summary, helping you to see the decision-making logic at a glance.
However, Anthropic has mentioned that, in most cases, the models’ thought processes are extremely simple and efficient, and presenting them directly is completely viable. This summarization feature, for the most part, is just an extra perk.
Notably, in performing multi-step tasks that require AI to complete autonomously, their tendency to exploit loopholes or shortcuts to achieve goals has decreased by <s
The Dynamic Duo: Each with Its Strengths
The release of Claude Opus 4 and Claude Sonnet 4 provides a diverse set of tools that cater to individual user needs.
Claude Opus 4, the undisputed performance champion, is purpose-built for tackling the most demanding tasks and hard problems. Its programming capabilities are top-notch, scoring an industry-leading 72.5% on the SWE-bench (Software Engineering Benchmark) and 43.2% on the Terminal-bench (Terminal Operation Benchmark).
If you need AI to deeply engage in professional-level coding, intricate scientific research, meticulous legal document analysis, or strategic planning that demands robust, logical reasoning, then Opus 4 is the best option.
Claude Sonnet 4, on the other hand, excels as a master of balance between performance and efficiency. It also scored an impressive 72.7% on the SWE-bench, a significant advancement over the previous Sonnet 3.7 model. It responds with more precision to your instructions and provides high-quality content.
Sonnet 4 is an appealing, intelligent choice for users and businesses that need AI to deliver professional and reliable answers and assistance in their daily operations while balancing operational cost efficiency and responsiveness.
Claude Code: The Official Version Launches
Besides the stunning model upgrades, Anthropic has a long-awaited gift for developers around the world: the official version of Claude Code has been released! This isn’t just a simple utility to help you complete a few lines of code; the goal is to be your true smart programming co-pilot.
Claude Code assists in grasping, effortlessly browsing, and precisely editing vast code repositories, enabling you to confidently delegate arduous tasks to AI, like fixing complex bugs, implementing new functionality from scratch, massive code refactoring, writing comprehensive test cases, and even managing complicated modifications across multiple files.
Anthropic has also released the scalable Claude Code SDK, meaning that able developers and teams can make their custom AI agents and applications using its core agent.
Sonnet 4 is Now Free
Great news: starting today, all Anthropic’s paid users (including Pro, Max, Team, and Enterprise packages) are able to instantly try out the full power of the two amazing devices, Claude Opus 4 and Claude Sonnet 4.
Anthropic, however, also cares about developers and regular users, so even free users can now use Claude Sonnet 4.
The developer community is getting the ultimate “gift pack” this time: Anthropic’s APIs have also received a substantial update, introducing the code execution tool, MCP connector, and file API, accompanied by the helpful characteristic of a cache for prompts lasting one hour. These features will pump up developers’ ability to build powerful and intelligent AI apps.
Regarding the pricing, , Anthropic has kept its consistent transparency and sincerity, with API pricing consistent with those of earlier products, aiming to make it affordable to more people:
Claude Opus 4: Input at $15/million tokens, output at $75/million tokens.
Claude Sonnet 4: Input at $3/million tokens, output at $15/million tokens.
As the AI’s “technology tree” grows, the arrival of “loaded” machines like Claude 4 continues. It seems that every model gets better. Claude 4 can now play Pokémon for 24 hours continuously. Who knows, maybe AI will become a gamer someday (doge).
Original article, Author: Tobias. If you wish to reprint this article, please indicate the source:https://aicnbc.com/1051.html
