Anthropic has just introduced two new versions of Claude, its renowned conversational AI. The most notable is Claude Opus 4, its most advanced coding model to date and an expert in solving complex problems. On the other hand, we find Claude Sonnet 4, the evolution of the model Sonnet 3.7, which boasts great coding and reasoning capabilities, aimed at everyday use.
Both AIs are hybrid models, meaning they offer the option to activate or deactivate their advanced reasoning capabilities, depending on the complexity of the task at hand. It is worth noting that Claude Sonnet 3.7, introduced in February of this year, was the first hybrid model created by Anthropic.
Claude Opus 4 and Claude Sonnet 4 are now available in the Pro, Max, Team, and Enterprise plans of Claude, while the free plan includes only Sonnet 4. Additionally, both AIs are included in the Anthropic API, Amazon Bedrock, and Vertex AI from Google Cloud.
The Claude 4 models are the most powerful of the Claude family to date, setting new standards in coding, deep reasoning, and agentic AI. Both AIs have also been optimized to avoid using shortcuts or loopholes to complete tasks, with a 65% reduction in the likelihood of exhibiting such behavior.
It is noteworthy that these models can use tools, such as web search, while applying their extended reasoning. This allows them to alternate between reasoning and tool use to provide better results and responses. However, at present, this capability is still in the beta phase.
As explained by Anthropic, both Claude Opus 4 and Claude Sonnet 4, “can use tools in parallel, follow instructions more precisely, and when developers give them access to local files, they showcase significantly improved memory capabilities, extracting and storing key data to maintain continuity and generate tacit knowledge over time.”
Claude Opus 4 is designed to offer sustained performance in complex and long-duration tasks, which demand concentrated effort and the execution of thousands of steps. This AI has the ability to work continuously for hours, marking a turning point in the Claude family of models.
Moreover, it features a remarkably superior memory compared to previous models. This, combined with the aforementioned skills, has helped improve its gaming experience while playing Pokémon. An experiment already tested with Claude 3.7 Sonnet.
Another noteworthy aspect is its potential to drive AI agent workflows, significantly enhancing the capacity of these agents. All these characteristics make it a tool geared toward tackling highly complex tasks in various fields, such as programming, research, science, or writing, among others.
Anthropic defines Claude Opus 4 as “the world’s best coding model“, and to justify these claims, they have shared some of the results achieved by the model in different tests. For instance, Claude Opus 4 scored 72.5% on SWE-bench and 43.2% on Terminal-bench.
In addition, the developer has also shared the experiences of companies such as Cognition, Cursor, Replit, Block, or Rakuten using Claude Opus 4 for various tasks.
For its part, Claude Sonnet 4 is less advanced than Claude Opus 4, but it is the most cutting-edge model of the Sonnet range. Its deep reasoning and coding capabilities are highly precise and present significant improvements in its ability to respond to given instructions.
According to data presented by Anthropic, Claude Sonnet 4 scored 72.7% on coding SWE-bench tests. “The model balances performance and efficiency for internal and external use cases, with greater manageability for increased control over deployments. While it does not match Opus 4 in most domains, it offers an optimal combination of capability and practicality,” explains the developer.
Whereas the Opus 4 model is intended for highly professional use cases, Sonnet 4 “brings cutting-edge performance to everyday use cases as an instant upgrade from Sonnet 3.7.”
The AI developer has leveraged the launch of its new models to introduce other improvements:
Photo: Anthropic
Your email address will not be published. Required fields are marked *
Δ