Back to rankings

Lovable: Fable 5.1 Powers App Building

LovableView original changelog

Lovable's core app-building agent now runs on Fable 5.1, Anthropic's newest model, replacing Fable 5 as the model Lovable reaches for on its hardest tasks. Lovable's internal benchmarks show the biggest gains land in iterative code fixing (up to 17.2% better at high reasoning effort) rather than fresh builds, alongside a 31 to 39% drop in cost for that work. Lovable is currently A/B testing the rollout rather than switching every builder over at once.

Key Takeaways

  • Lovable's core app-building agent now runs on Fable 5.1, Anthropic's newest model, in place of Fable 5.
  • The biggest gains land in iterative code fixing, the category Lovable says represents the majority of real builder work, improving up to 17.2% at high reasoning effort.
  • Those iterative-fixing gains come alongside a 31 to 39% drop in cost per task depending on reasoning effort.
  • Fresh, from-scratch builds actually scored slightly lower than Fable 5 at high effort, showing the upgrade is not a universal improvement.
  • Fable 5.1 is more disciplined about self-verification, opening the running app in a browser to confirm it works before marking a task done.
  • Lovable is currently A/B testing the switch, routing stuck sessions and longer, more complex builds to Fable 5.1 rather than a full rollout.

Lovable's Coding Agent Moves to Fable 5.1

Lovable announced that its core app-building agent now runs on Fable 5.1, Anthropic's newest model, replacing Fable 5 as the model Lovable reaches for on its hardest tasks. The upgrade was detailed in a blog post from the company, alongside internal benchmark results comparing the two model versions.

Where Fable 5.1 Gains the Most

According to Lovable's internal benchmark suite, which scores models on building apps from scratch, fixing and evolving existing codebases, and visual design quality, Fable 5.1's biggest gains are not in fresh builds. New builds actually scored slightly lower than Fable 5 at high reasoning effort (down 2.7%). Instead, Fable 5.1 pulled ahead sharply on iterative code fixing, the category Lovable says represents the largest share of what its builders actually do day to day: improving apps that are already live. Lovable measured gains of up to 17.2% on that benchmark at high reasoning effort, while simultaneously cutting the cost of those tasks by roughly 31 to 39%, depending on effort level. Visual design quality also improved modestly, up 3.5% at high effort.

What's Different Under the Hood

Lovable attributes the jump on existing-codebase work to three factors: Fable 5.1 is more disciplined about self-verification, opening the running app in a browser to confirm it actually works before marking a task complete; it shows a clearer advantage specifically when extending live, already-complex apps rather than starting fresh; and its design judgment sharpens further as it is given more reasoning effort to work with.

Lovable co-founder and CTO Fabian Hedin said the model is "just more careful, checking its own work and opening the app to make sure it actually runs before calling anything done," adding that the result "looks and works the way you pictured it."

Rollout

Lovable said it is currently A/B testing the switch, routing sessions that get stuck, along with longer and more complex build requests, to Fable 5.1, rather than flipping every builder over at once.