Claude (AI model)

2025

Anthropic introduced Claude 3.7 Sonnet

The official release took place on February 24, 2025, but the long-awaited search agent was never presented, but they presented an adapted reasoning model.

Anthropic shifts priorities towards programming and user agents - this is what most of the presentation was built on.

Anthropic introduced Claude Code, a command-line coding tool that is in limited preliminary research. This tool can search, read and edit code, write and run tests, and interact with GitHub, supporting test-based development, debugging, and refactoring.

One of the key features is the hybrid mode, allowing users to choose between the standard mode for quick responses and the advanced mode for in-depth, step-by-step thinking.

The new version demonstrates the best performance in solving multi-step problems, including mathematics, financial analysis, legal inquiries and even the passage of complex game scenarios. This makes it especially useful (potentially, in practice, not everything is as good as in presentations) for business applications and scientific research, where high reliability and transparency of the model is required.

It is stated that the extended mode of thinking significantly improves results in mathematics, programming and science.

Anthropic's internal tests showed that the new model reduces the number of unreasonable failures in responses by about 45% compared to the previous version. Reduced generation errors (hallucinations).

The model supports a context of up to 200 thousand tokens and can generate up to 128 thousand output tokens.

According to tests from Anthropic looks impressive, but not revolutionary, wrote Spydell Finance. Comparable to GPT o1/o3-high, DeepSeek R1 and Grok 3 depending on tests, but integrally claims the world's best LLM, competing with Grok 3.

There was no qualitative breakthrough, it is fair to talk about equalizing competition.

Thus, as of February 25, there are only 5 advanced reasoning models in the world:

• Claude 3.7 Sonnet Thinking
• Grok 3 Reasoner
• GPT o1/o3-high
• DeepSeek R1
• Gemini 2.0 Thinking mode.

Anthropic is expected to break into the leadership group, but more complete tests are needed.

For professional users, direct access to paid models for fine tuning is a priority, but most of the current tasks are solved without in-depth modifications of the models.

The pace of innovation is prohibitive: DeepSeek R1 in mid-January, GPT-o3 in early February, a week later Gemini 2.0, the other day Elon Musk pleased with Grok 3, and now Claude 3.7 Sonnet.

GPT-4o and other neural networks do not cope with most programming tasks - OpenAI study

Large language models (LLMs) greatly simplify and speed up the writing of program code, but they are not able to independently cope with most programming tasks. This is stated in the OpenAI study, the results of which were published in mid-February 2025. Read more here.

2024: Claude 3 Model Announcement

On March 4, 2024, Anthropic, founded by immigrants from OpenAI, announced models of artificial intelligence of the Claude 3 family. They are said to surpass the counterparts of both OpenAI itself and Google.

The family includes three solutions: Claude 3 Haiku, Claude 3 Sonnet and Claude 3 Opus. Depending on the implementation, these models allow you to choose the optimal balance of AI performance and cost for a specific application. Opus and Sonnet are available for use on the platform claude.ai. In addition, they can be accessed through a specialized program interface (API).

𝓲

Фото: Anthropic

Anthropic claims that its Claude 3 Opus model is superior to GPT-4 and Gemini in solving mathematical problems, computer coding, general knowledge and other fields. Moreover, as noted, Opus exhibits "an almost human level of understanding" and fluency in answering difficult questions. In addition, all models of the Claude 3 family show advanced capabilities for analyzing and predicting, creating detailed content, generating code and communicating in various languages, including Spanish, Japanese and French.

As of early March 2024, Claude 3 Haiku, according to Anthropic, is the fastest and most economical AI model on the market in its category. Claude 3 models have advanced machine vision capabilities: they can handle a wide range of graphic materials, including photos, diagrams and technical data. Claude 3 models can also be used in interactive chats with users.

In a test that requires master's level reasoning, the Claude 3 Opus model showed a result of 50.4% versus 35.7% for GPT-4. And for the Claude 3 Sonnet version, this figure was 40.4%.^[1]

Notes