Anthropic upgrades Claude Opus to version 4.8 at the same price.

Anthropic upgrades Claude Opus to version 4.8, boosting coding and reasoning at the same price while adding an effort setting and Claude Mythos Preview.

Claude Opus 4.8 succeeds Opus 4.7, showcasing demonstrated improvements across all coding, reasoning, agentic tasks, and intellectual work tests, all at the same price as the previous version: $5 per million input tokens and $25 per million output tokens.

Emphasis is placed on the model's honesty. According to the publisher, Opus 4.8 more readily signals its uncertainties and makes fewer unsubstantiated conclusions: it reportedly allows approximately four times fewer defects to go unnoticed in the code it produces. The in-house alignment evaluation also reports problematic behaviors, such as deception or complacency in the face of misuse, significantly less frequent than on Opus 4.7.

The release comes with several new features. An "effort setting" appears on claude.ai, allowing users to balance response speed and depth of thought, available on all plans. The fast mode, which runs at 2.5 times the speed, is now three times cheaper than on previous models. For developers, the Messages API now accepts system instructions mid-conversation, without breaking the prompt cache. Finally, the version is accompanied by dynamic workflows in Claude Code, a feature covered in a separate news item.

The publisher also mentions an even more powerful class of models, a preview of which, named Claude Mythos Preview, is already being tested by some organizations for cybersecurity uses, while awaiting reinforced safeguards.