Anthropic launches Opus 4.8 and new Dynamic Workflows tool

The accelerated release comes at a time of increasingly fierce competition among leading AI providers. OpenAI recently introduced GPT-5.5 and new Codex capabilities, and Google Gemini 3.5 Flash. Anthropic clearly doesn't want to be left behind in the race for the most powerful models.

Anthropic is not just emphasizing better results on standard benchmark tests with Opus 4.8, but also its improved handling of incomplete or uncertain data. According to the company, the model is much more willing to point out missing information, uncertainty in results, or potential analysis problems, rather than giving confident answers without sufficient basis.

This is an important change, as many users of generative artificial intelligence have often criticized so-called hallucinations - cases where the model gives a convincing but incorrect answer.

Among the first users is investment giant Bridgewater Associates, who pointed out that Opus 4.8 significantly more often points out problems in input data and analysis results that other models would often overlook.

Along with the model, Anthropic also introduced a new feature called “Dynamic Workflows,” which is currently available in a research version. It is a system that allows complex tasks to be divided among a large number of parallel UI agents, or subagents. Instead of one model solving the entire task, it can divide the work into multiple parts that the agents execute simultaneously.

Anthropic cites migrations of large software systems as an example, where Claude Code, together with Opus 4.8, can manage projects involving hundreds of thousands of lines of code.

This is another step towards agent-based artificial intelligence, where models no longer just answer questions, but independently carry out larger projects.

Despite the new release, Anthropic has yet to publicly launch its most advanced model, the Mythos. The company launched a limited trial a month ago, but concerns about cybersecurity and potential abuse have led Anthropic to implement additional safeguards and delay a wider release.

In a new announcement, the company suggests that the Mythos could soon become available to a wider range of users. According to them, the development of security measures is progressing rapidly and Mythos-class models could be available in the coming weeks.

The launch of Opus 4.8 confirms a broader trend in the AI industry. The competition between OpenAI, Google, and Anthropic is no longer focused solely on the quality of answers, but on the ability of models to independently perform complex tasks.

New models increasingly act as digital collaborators who can plan work, use tools, verify results, and collaborate with other agents.