Anthropic’s Claude Opus 4.8: A More Honest AI Model
Anthropic has released Claude Opus 4.8, its most advanced and honest AI model to date, with significant improvements in reliability and accuracy. The company also teased the upcoming Mytheos class models, which have already identified over 10,000 software vulnerabilities through Project Glasswing.
Key Updates:
- Honesty and Reliability: Opus 4.8 is four times less likely to introduce code flaws compared to its predecessor, Opus 4.7. It is designed to be more transparent about uncertainties and catch its own mistakes.
- Benchmark Improvements: The model outperforms Opus 4.7 across various benchmarks, including agentic coding, multidisciplinary reasoning, and knowledge work.
- Alignment Assessment: Opus 4.8 demonstrates enhanced prosocial traits, such as supporting user autonomy and acting in users’ best interests, with reduced misaligned behaviors like deception.
- Practical Applications: Early testers from companies like Cognition, Cursor, Harvey, Databricks, Thomson Reuters, and Hebbia have praised the model’s improvements in code generation, tool usage, legal analysis, and financial document analysis.
Quote:
"Anthropic has released Claude Opus 4.8… with sharper judgement and four times fewer uncaught code flaws."
Details:
- Availability: Opus 4.8 is available immediately at the same pricing as its predecessor.
- Integration: It is rolling out across all Anthropic products, including claude.ai, Claude Code, and the API.
- Series H Round: Anthropic announced a $65 billion Series H round at a $965 billion post-money valuation.