Claude Fable 5 arrives on GitLab Duo — multi-day agent 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 since June 9, 2026. This is not a routine update: developers, platform engineers, and technical leads get a tool capable of independently completing multi-stage, goal-directed tasks that previous models could not handle without constant manual intervention. GitLab Duo features — Agentic Chat and basic agents — receive from Fable 5 fundamentally expanded capabilities.

Accurate Result on First Attempt

Claude Fable 5 has significantly improved first-attempt accuracy on complex, well-formulated tasks. Early testers reported: systems that previously took days of iterations are now ready in a single pass without rework. In GitLab Duo Agentic Chat this means fewer back-and-forth dialogs.

If you describe a task precisely, the model is more likely to hit the mark and return working code on the first try. The effect is especially noticeable where the cost of iterations is high: refactoring multiple files, investigating production incidents, describing infrastructure as code. Fable 5 is also significantly more accurate than its predecessors at parsing dense technical images, web application screenshots, and detailed diagrams.

At the same time, the model often spends fewer output tokens than previous versions — despite higher-quality results.

Agentic Workflows Without Manual Stops

The most significant change in Fable 5 is a long autonomy horizon. The model maintains a productive course in multi-day goal-directed runs, following instructions across millions of tokens without losing task focus. Previous models required periodic intervention — Fable 5 behaves differently. For teams on GitLab Duo Agent Platform, this means: workflows that previously required manual checkpoints or repeated instructions now complete independently. Fable 5 corrects its own errors through verification loops, maintains task focus, and is significantly more reliable at managing parallel sub-agents — including in workflows spanning multiple repositories or services.

"Teams seeing the best results direct

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

For engineering leaders evaluating AI deployment, this has concrete operational consequences: it reduces the cost of human oversight for each agent run. Teams can assign agents more complex problems, review results asynchronously, and trust that the output reflects actual progress, not work simulation.

More Bugs — Faster Incident Triage

Bug-finding completeness has grown noticeably compared to previous models. Incident triage, root cause analysis, and repository history investigation — all improved. When reviewing merge requests through GitLab Duo Agent Platform, Fable 5 catches issues that previous models missed. Where improvements are most noticeable:

• Incident triage and identification of incident root causes
• Repository history analysis when investigating production failures
• Code review with specific, actionable comments instead of general remarks
• Consistent identification of edge cases when checking pull requests
• Deeper code path tracing when debugging errors

For teams managing CI/CD, this has concrete consequences: fewer defects in production and accelerated mean time to recovery after failures.

What This Means

Claude Fable 5 shifts the bar for what can be delegated to an agent without constant oversight. GitLab became one of the first platforms where this capability is available to developers on any tier. Anthropic's recommendation is unambiguous: start not with routine tasks, but with the most complex unsolved problem in your repository — that's where Fable 5's capabilities shine in full force.