Anthropic Promises Its New Claude AI Models Will Be Less Likely to Try to Trick You
While it doesn’t have the same notoriety as ChatGPT or Google Gemini , Anthropic’s Claude AI bot continues to improve and innovate. All-new Claude 4 models are now available, promising improvements in coding, reasoning, accuracy, and the ability to manage long tasks independently.
There are two new models, the Claude Opus 4 and Claude Sonnet 4, and Anthropic says they both “set new standards” for what to expect from AI. Coding is a big focus, and the models are said to have achieved the highest scores to date in two widely used AI coding benchmarking tools, SWE-bench and Terminal-bench. The Claude 4 models can actually run for hours on projects without any user input, Anthropic says .
The updated models are better at handling more steps in more complex tasks, debugging their own work, and solving complex problems along the way. They should also follow user instructions more accurately and produce output that looks better and works more reliably. Anthropic cites partners like GitHub, Cursor, and Rakuten to explain how much of a step up these models represent.
In addition to generating and analyzing code, the models also bring with them advanced thinking, the ability to work on multiple tasks in parallel, and improved memory. They are better at integrating web searches as needed, as well as checking for supporting information and making sure they are on the right track with their answers.
Also new are “summary summaries” that give deeper insight into how Claude 4 came to its conclusions, and an “extended thinking” feature launched in beta that lets you force the AI bot to spend more time thinking about its answers.
Anthropic is now making its Claude Code toolset available to a wider audience, another step toward agent-based AI that can operate autonomously, without constant assistance from human users. A demo video shows Claude 4 models scraping research papers from the web, creating an online ordering system, and extracting information from documents to create achievable tasks.
Claude 4 is now available (but you’ll have to pay extra for the more advanced model)
The Claude Sonnet 4, which is faster and doesn’t have the same capabilities in terms of thinking, coding, and memory, is now available to all Claude users. The more advanced Claude Opus 4, which also includes additional tools and integrations, is available to users of any of Anthropic’s paid subscriptions.
The road to the release of these Claude 4 models hasn’t been smooth: Anthropic claims its security advisory partner warned against releasing earlier versions of the models due to their propensity for “intrigue and deception.” Those issues now appear to have been resolved, but it’s a reminder that as AI models become more powerful, they should also come with better guardrails and additional security features.
I’m not a real programmer, so I can’t comment with complete confidence on the major updates included in Claude 4, but I was able to test the advanced reasoning and thinking capabilities of Claude Sonnet 4 and Claude Opus 4. These capabilities are not easy to quantify or measure, but all the responses I received were well written and well presented, and as far as I could tell, contained accurate information with online citations.
Honestly, I’m always a little stuck when it comes to how to fully utilize AI chatbots and their latest updates. They can definitely save time when doing certain web searches and researching topics online, but I don’t fully trust the results or the AI’s ability to decide what’s relevant and what’s not — I’d still rather read and summarize myself, even if it’s slower.
Maybe I should start a coding project and see how far I can get on vibrations alone. I asked Claude Opus 4 to create a simple HTML time tracker for me that I could run in a browser tab to make sure I wasn’t wasting too much time on distractions throughout the day. It did the job in a couple of minutes and spun out something that worked well, close to my instructions. While it worked fine, Claude 4 reported a few errors along the way that I certainly didn’t understand—I guess I can ask the AI about that.
Anthropic isn’t the only AI company showing off new models. At Google I/O 2025 earlier this week, the company unveiled improved coding assistance and thought summaries in Gemini, following up on the announcement of its best AI models a few weeks ago. Meanwhile, OpenAI has been testing its GPT-4.5 model since February , boasting improvements in coding and problem solving.