Anthropic unveiled Claude Sonnet 5 on Tuesday, calling it the firm’s "most agentic Sonnet model yet." The label reflects a new design focus: rather than merely answering questions, the model can chart a course of action, invoke external tools such as web browsers and command-line terminals, and carry out assignments with far less step‑by‑step prompting.
According to the company, Sonnet 5 can "make plans, use tools like browsers and terminals, and run autonomously at a level that, just a few months ago, required larger and more expensive models." The upgrade targets professional workloads—particularly coding and everyday office tasks—where users demand not just advice but concrete execution.
Performance metrics back up the claim. In the Agentic Coding segment of the Terminal‑bench 2.1 suite, Sonnet 5 achieved an 80.5% success rate, a sizable jump from the 67% logged by its predecessor, Sonnet 4.6. Anthropic says the improvement narrows the gap between cost‑effective models and the high‑end offerings that previously dominated complex tool‑use scenarios.
Access to the new model is broad. Anthropic has made Sonnet 5 the default engine for every user tier, from the free plan to the Pro subscription. It also rolls out to Max, Team, and Enterprise customers, and is embedded in Claude Code as well as the Claude Platform overall. The decision to democratize the technology underscores the company’s belief that agentic capabilities will soon become a baseline expectation, not a premium add‑on.
The launch arrives amid a broader industry pivot. Google’s Gemini Spark, billed as a 24/7 agentic personal assistant, debuted earlier this month, signaling that major players are racing to embed planning and tool‑use into their AI products. At the same time, Anthropic’s own Claude Fable 5 and Mythos 5 have attracted regulatory scrutiny, while OpenAI’s GPT‑5.6 remains under review by U.S. authorities.
Analysts interpret the shift as a move away from the traditional chatbot arms race, which centered on conversational fluency, toward a competition focused on execution. "The next stage of the AI war will not be won by the chatbot that gives the neatest answer," an Anthropic blog post warned. "It will be won by the assistant that can take a messy task, keep track of the plan, and actually get something useful done."
For developers, the change could mean more reliable code generation without the need for extensive prompt engineering. Business users may soon rely on AI to draft reports, schedule meetings, or even troubleshoot software—tasks that previously required human oversight at every step.
While the rollout is enthusiastic, Anthropic acknowledges that autonomous behavior raises safety considerations. The company says it continues to refine guardrails and monitoring tools to prevent misuse, especially as the model gains broader reach across free and paid tiers.
In short, Claude Sonnet 5 represents a tangible leap toward AI assistants that act more like collaborators than mere information sources. Whether the market will quickly adopt this new paradigm remains to be seen, but the signal from Anthropic is clear: the future of generative AI lies in doing, not just talking.
Cet article a été rédigé avec l'assistance de l'IA.
News Factory APP - actualités agentiques pour booster votre SEO et AEO.