Claude 3.7 Sonnet – the revolution has not happened

      Anthropic has introduced a new version of its AI, Claude 3.7 Sonnet, a model with a hybrid mode of operation combining instant answers and deep step–by-step reflection. Whether these updates will be able to change the balance of power in the world of large language models, or whether this is just a marketing ploy to level the competition, the IT World is investigating.

       On February 24, Anthropic officially

      unveiled the Claude 3.7 Sonnet, its most advanced model to date, which has already caused a stir among experts. The novelty is able to work in two modes: one for quick and concise answers, the other for in–depth, detailed analysis, where the algorithm literally "reflects" on the task. This solution, according to the company, allows the model to better handle multi-step tasks, whether it's mathematical calculations, financial analysis, legal inquiries, or even complex game scenarios.

       The key argument in favor of the new release was improvements in the logic and accuracy of the answers. According to internal tests by Anthropic, Claude 3.7 Sonnet reduces the number of unjustified rejections by about 45% and significantly reduces the number of errors (so-called "hallucinations") during text generation. The model is capable of processing the context of up to 200 thousand tokens, and the volume of generated responses can reach 128 thousand tokens. Such indicators are undoubtedly of interest, although comparisons with the GPT-o1/o3-high, DeepSeek R1 and Grok 3 models show that the performance gap is not so dramatic yet – rather, the Anthropic is firmly entrenched in the top five along with the Grok 3 Reasoner, GPT o1/o3-high, DeepSeek R1 and Gemini 2.0 Thinking Mode.

      

       At the same time, along with Claude 3.7 Sonnet, the company introduced a new development – Claude Code. It is a command-line tool that is capable not only of reading and editing code, but also of writing and running tests, as well as interacting with GitHub. According to Anthropic, Claude Code performs tasks that require developers to perform about 45 minutes of manual labor in one pass. In an environment where time is money, such a solution can be a real gift for specialists in the field of programming and development. Despite the fact that Claude Code is currently available only in a limited research mode, the first results are encouraging: automated programming, in fact, has already ceased to be fiction.

       Interestingly, the new model is available in all tariff plans – from free to corporate, as well as through the API Anthropic, Amazon Bedrock and Google Cloud Vertex AI. The cost of use remains the same: $ 3 per million input tokens and $15 per million output tokens, including "reflection tokens". However, users of the free tariff are deprived of access to the extended mode of thinking – and here, of course, you can tease conditional "free users", because you have to pay for advanced features.

       As a result, Claude 3.7 Sonnet can be called an important step in the evolution of large language models. Yes, there has not been a technological breakthrough yet, and the model is more likely to catch up with competitors than radically change the rules of the game. But considering that in today's world of advanced AI, there are only five solutions that work, and a lot depends on it, such a step looks very significant. It is especially interesting to see how Anthropic relies on hybrid thinking and agent–based systems, an approach that is already well-known among OpenAI and xAI.

       Observing the pace of innovation – from the introduction of DeepSeek R1 in mid-January to the release of GPT–o3 in early February and the subsequent announcement of Grok 3 - it becomes clear that the market for large language models is rapidly growing and changing. And the upcoming release of GPT-4.5 Orion may again reconsider the balance of power in this niche. In this race, the Anthropic is confidently taking its place with its new model, but the real battle for leadership is just beginning.

       Ultimately, Claude 3.7 Sonnet is not so much a revolution as an evolution in which every little thing matters. Perhaps a real breakthrough idea is already on the horizon, but in the meantime, the world is watching technology bring us step by step closer to a new level of automation and intelligent capabilities.

      Read also

       About the API as the leading trend in fintech development. IT-World Analytics

      In 2026, the Central Bank of the Russian Federation will oblige all major Russian banks, insurance companies and brokers to use Open Application Programming interfaces. Starting in 2027, this responsibility will fall on MFIs, depositories, financial platforms, and CFA issuance operators. Why the Central Bank of the Russian Federation is transferring financial market players to the Open API and what will happen next, the IT-World figured out.