Anthropic has announced Claude Opus 4.8, its latest flagship AI model, which the company says improves coding, reasoning, speed, and reliability. The model is also designed to better recognize uncertainty and identify potential issues before presenting answers to users.
Claude Opus 4.8 focuses on reliability and coding performance
Early testers found Claude Opus 4.8 more reliable during complex tasks and less likely to make unsupported claims.
According to Anthropic, the model is better at identifying uncertainty in its own work and more likely to flag potential flaws in generated code instead of overlooking them.
Anthropic also highlighted improvements in alignment and safety, claiming the model performs better on measures related to user autonomy and acting in a user’s best interests while reducing deceptive behavior compared to previous versions.
On benchmarks shared by the company, Claude Opus 4.8 scored 69.2% on SWE-Bench Pro, placing ahead of GPT-5.5 and Gemini 3.1 Pro on that benchmark.
The model’s fast mode now runs 2.5 times faster than before while costing roughly one-third as much as earlier versions.
Anthropic is also adding new tools and workflow features
Alongside the new model, Anthropic announced several platform updates.
The new additions include:
- Dynamic Workflows for handling larger coding projects through parallel AI agents
- Effort Control settings that let users choose how much reasoning Claude uses
- Messages API improvements that allow developers to modify instructions during tasks
Dynamic Workflows is launching in research preview for Claude Code users on Enterprise, Team, and Max plans.
Anthropic says Claude Opus 4.8 is available immediately, and pricing remains unchanged from Claude Opus 4.7.
The company also revealed that it is working on future models with similar capabilities at lower costs, alongside more advanced systems beyond Opus. Anthropic says its experimental Claude Mythos model is currently being tested with a limited number of organizations and could become more widely available in the coming weeks.
Do you care more about AI models becoming faster and cheaper, or would you rather see improvements in accuracy and reliability? Share your thoughts in the comments below.
You might also like:
