Claude Opus 4.8 on GitLab: Complex agentic work, less disruption (opens in new tab)

Claude Opus 4.8 is now available in GitLab Duo Agent Platform for complex, multi-step coding and agentic workflows. The model is designed to follow instructions more precisely over extended autonomous runs, reducing the need for human correction and intervention. It also supports mid-conversation system-prompt updates without invalidating the prompt cache, making it better suited to changing contexts.

Improved Long-Horizon Agentic Execution

• Handles complex workflows involving multiple tools and extended sequences.
• Provides more comprehensive planning and reasoning.
• Produces more accurate end states with fewer interruptions or redirects.
• Interprets instructions more precisely than previous models.
• Supports professional tasks beyond coding, including:
  • Document drafting
  • Data analysis
  • Structured knowledge work

Mid-Conversation System Prompts

• System instructions can change during an active session.
• Prompt-cache validity is preserved when context changes.
• This is useful when:
  • Files change on disk
  • Token budgets shift
  • User context is updated
  • Asynchronous information arrives

Availability and Pricing

• Available through model selection in Agentic Chat and across GitLab Duo agent workflows.
• Runs on GitLab Credits.
• GitLab Premium and Ultimate subscribers can use included credits.
• New users can access it through a GitLab Duo Agent Platform trial or the GitLab Free tier.

Teams seeking more reliable autonomous coding and documentation workflows can use Opus 4.8 in GitLab Duo to reduce manual oversight while agents complete longer, more complex tasks.