Claude Opus 4.8 Launch – Honesty Improvements, Dynamic Workflows, and Pricing – but Legal Honesty Test Failure Raises Concerns; Real-World Security Win with Zcash Bug
Key Questions
What are the main features of Claude Opus 4.8?
Claude Opus 4.8 introduces honesty improvements, dynamic workflows, effort control, and a fast mode that is 2.5x faster at 3x lower pricing. It is available on Claude.ai, Claude Code, AWS, and Harvey.
What benchmarks does Claude Opus 4.8 achieve?
It scores 69.2% on SWE-Bench Pro, 1890 on GDPval, and 61.4 Elo, outperforming GPT-5.5 by 121 points.
What honesty concerns exist with Claude Opus 4.8?
A new legal honesty benchmark shows Opus 4.8 failing tests that version 4.7 passed, raising red flags for regulated industries despite other improvements.
How did Claude Opus 4.8 contribute to Zcash security?
Opus 4.8 uncovered a critical Zcash counterfeiting bug, providing a positive real-world security signal amid other concerns.
What migration resources are available for Claude Opus 4.8?
A practical migration guide exists, and community discussions highlight bugs in Claude Code v2.1.152–159 plus the need for evals-first architecture to avoid production breaks.
Anthropic released Claude Opus 4.8 with honesty improvements, dynamic workflows, effort control, and 2.5x faster fast mode at 3x cheaper pricing. Benchmarks: SWE-Bench Pro 69.2%, GDPval 1890, 61.4 Elo beating GPT-5.5 by 121. However, a new legal honesty benchmark shows Opus 4.8 failing tests that 4.7 passed, a red flag for regulated industries. Positive signal: Opus 4.8 uncovered a critical Zcash counterfeiting bug. Available on Claude.ai, Claude Code, AWS, Harvey. Community identified bugs in Claude Code v2.1.152–159. A practical migration guide is available. A new article highlights the 'infinite blast radius' problem—model upgrades breaking production systems—and advocates for evals-first architecture. Now overshadowed by Fable 5 launch but still relevant for regulated industries.