Claude Fable 5 Arrives on GitLab Duo: Multi-Day Agentic Tasks Without Manual Stops

Claude Fable 5, Anthropic's flagship Mythos-class model, is now available on GitLab Duo Agent Platform across all tiers and all deployment models as of June 9, 2026. This is not a routine update: developers, platform engineers and technical leaders get a tool that can complete multi-step, goal-driven tasks on its own, tasks that earlier models could not handle without constant manual intervention. GitLab Duo features, Agentic Chat and the core agents, gain substantially expanded capabilities from Fable 5.

Accurate results on the first attempt

Claude Fable 5 has noticeably improved first-attempt accuracy on complex, well-specified tasks. Early testers reported that systems which previously took days of iteration are now done in a single pass without rework. In GitLab Duo Agentic Chat, this means fewer back-and-forth exchanges. If you describe the task precisely, the model is more likely to hit the target and return working code the first time.

The effect is most visible where iterations are expensive: refactoring across multiple files, investigating production incidents, and describing infrastructure as code. Fable 5 also reads dense technical images, web application screenshots and detailed diagrams significantly more accurately than its predecessors. The model often spends fewer output tokens than previous versions, despite producing better results.

Agentic workflows without manual stops

The most significant change in Fable 5 is its long autonomy horizon. The model stays on a productive course through multi-day goal-driven runs, following instructions across millions of tokens without losing focus on the task. Previous models needed periodic intervention; Fable 5 behaves differently.

For teams on GitLab Duo Agent Platform, this means workflows that used to require manual checkpoints or repeated instructions now reach the end on their own. Fable 5 corrects its own mistakes through verification loops, stays focused on the task, and manages parallel subagents far more reliably, including in workflows that span several repositories or services.

"The teams seeing the best results point

Fable 5 at tasks they considered impossible for previous models," from Anthropic's internal testing.

For engineering leaders evaluating AI adoption, the practical consequence is concrete: the cost of human oversight for each agentic run goes down. Teams can assign agents harder problems, review results asynchronously, and trust that the output reflects real progress rather than an imitation of work.

More bugs found, faster incident analysis

Bug-detection thoroughness has grown noticeably compared with previous models. Outage triage, root-cause analysis and repository history investigation have all improved. When reviewing merge requests through GitLab Duo Agent Platform, Fable 5 spots problems that earlier models missed.

Where the improvements are most noticeable:

• Outage triage and identifying root causes of incidents
• Repository history analysis when investigating production failures
• Code review with specific, actionable comments instead of generic remarks
• Consistent detection of edge cases when checking pull requests
• Deeper tracing of code paths when analyzing errors

For teams running CI/CD, the concrete outcomes are fewer defects reaching production and a shorter mean time to recovery after failures.

What this means

Claude Fable 5 raises the bar for what you can hand to an agent without constant supervision. GitLab is among the first platforms where this capability is available to developers on any tier. Anthropic's recommendation is unambiguous: don't start with routine tasks, start with the hardest unresolved problem in your repository, because that is where Fable 5's capabilities show their full strength.