Daily AI Catchup
Model

Thinking Machines launches Inkling 2 Small, open weights soon

Thinking Machines released Inkling 2 Small, a mixture-of-experts model with 594 billion total parameters but only 17 billion active at once, keeping inference costs down. It accepts image inputs and can use visual feedback to iteratively refine websites, games, and 3D designs. It's available now through Tinker with closed weights, and full weights go Apache 2.0 licensed next week pending safety testing.

Why it matters

๐Ÿ’ป Developer ยท A large MoE model with visual feedback loops for iterative design work is worth testing once weights are open, especially for agentic UI-building tasks.

๐Ÿ“ฆ Product ยท Models that can look at their own output and fix visual bugs could make AI website and app builders noticeably more reliable, less 'close but broken.'

๐ŸŽจ Design ยท This is squarely aimed at your workflow: an AI that iterates on visual designs using what it actually sees, not just text descriptions.

๐Ÿ“ˆ Business ยท Open-weighting a 594B model under Apache 2.0 is a notable move that could undercut closed competitors on cost for teams building visual agents.

๐Ÿค” Just Curious ยท A new AI model can look at a half-finished website or game it's building, notice what's wrong, and fix it, similar to how a person would eyeball their own work.

Sources: Introducing Inkling 2 Small