Anthropic has announced Claude Opus 5, a new large language model the company describes as a “thoughtful and proactive model” that approaches the frontier intelligence of Claude Fable 5 while being offered at the same price as Opus 4.8.
Performance and positioning
According to the announcement, Claude Opus 5 currently leads the Artificial Analysis leaderboard, ranking ahead of Fable 5. Pricing matches Opus 4.8, and the model continues to offer a “fast mode” that costs twice the base model.
Example of proactive behavior
The release includes an anecdotal Frontier-Bench task in which Opus 5 was asked to write code to rebuild a machine part as a 3D FreeCAD model from a drawing. In that task the model was deliberately given no direct way to view the drawing; Opus 5 responded by implementing its own computer vision pipeline to extract geometry from the raw pixels and then reconstructed the full machine part. The example is presented as evidence of the model’s proactive problem-solving abilities.
Cybersecurity capabilities and limits
Anthropic states that, like its predecessor Opus 4.8, Opus 5 was intentionally not trained on cyber exploitation tasks. Nevertheless, the model has improved significantly on related tasks as it became more generally capable. Anthropic reports that Opus 5 comes close to Mythos 5 in finding cybersecurity vulnerabilities, but remains substantially behind Mythos 5 in exploiting those vulnerabilities—that is, converting findings into material cyber threats. The company explicitly notes it did not train the model on exploitation techniques.
Documentation
Anthropic has published a prompting guide for Claude Opus 5 to help users apply the model effectively.
Why this matters
Early benchmark success and the proactivity demonstrated in the release suggest Opus 5 advances trends toward more proactive and multimodal problem solving in large language models. At the same time, Anthropic’s decision to avoid targeted cyber training is intended to limit the model’s potential for misuse in cyber operations.
Note
The release post includes a brief personal anecdote by its author about being offline for part of the day; this appears as a stylistic element of the announcement rather than information about the model itself.
