Amazon-backed Anthropic has launched its latest language model, Claude 3.7 Sonnet, taking on the likes of ChatGPT and DeepSeek. Anthropic also took its first steps towards AI agents with a new tool for developers called Claude Code.
Notably, Claude has moved away from the traditional narrative of having a pre-trained model and an independent reasoning model. Instead, the Dario Amodei-led company has integrated the reasoning model alongside the frontier model, meaning that users can simply ask the chatbot when they want it to think longer before answering, essentially triggering the reasoning mode where it reflects on itself before answering a query. Claude says that triggering the reasoning mode (or as they call it, “extended thinking mode”) will help improve the model’s performance in maths, physics, following instructions and coding, among many other tasks.
What are reasoning models? Why is it a big deal?
Reasoning models are a different kind of LLM (the force behind various chatbots), designed to mimic human-like thinking to solve more complex problems. The first reasoning model, o1, was launched by OpenAI in September last year, and since then there has been a flurry of such launches from the likes of Google, DeepSeek and, more recently, Elon Musk’s Grok.
Anthropic says its reasoning model is optimised less for maths and computer science problems and more for solving real-world tasks. The company also claims that its new models will allow it to maintain its “coding leadership” in a highly competitive generative AI market.
It even presented some benchmarks in which Claude 3.7 Sonnet performed significantly better in terms of accuracy than its competitors, such as OpenAI o-3 Mini (High) and DeepSeek R1.
However, unlike other AI models such as Grok 3 Reasoning, DeepSeek R1 or o3 Mini (limited bandwidth), which are free to use, Claude 3.7 Sonnet’s extended reasoning mode will only be available to the company’s paying members. Meanwhile, free users will have access to the latest pre-trained model, Claude 3.7 Sonnet.
Anthropic developed its first-ever agentic tool:
Anthropic also announced its first agentic offering Claude Code – a command-line tool designed to allow developers to offload their extensive engineering tasks directly from their terminal – in a limited preview. Claude says it has improved the coding experience on its website, but its Github extension is also available on all plans, allowing developers to connect their coding repositories directly to the chatbot.
The company says Claude Code can “search and read code, edit files, write and run tests, commit and push code to GitHub, and use command-line tools” while keeping users in the loop every step of the way.
After Chinese AI startup DeepSeek disrupted the artificial intelligence (AI) industry in January 2025 with the release of its cost-efficient and powerful DeepSeek-R1 model, AI companies have been on an overdrive, intensifying efforts to innovate and compete. DeepSeek’s claim of training a ChatGPT-rivaling model for just $6 million—compared to the billions spent by U.S. giants like OpenAI and Meta—rattled Silicon Valley, triggering a reevaluation of resource-heavy strategies. It also sparked a stock market selloff that saw Nvidia lose nearly $600 billion in value in a single day.
