AI & ML · Jul 31
Inkling Small Hits 40, Nearly Matching Inkling 41 With 276B Parameters
Thinking Machines Lab’s smaller open weights model uses 276B total and 12B active MoE, while Inkling uses 975B total and 41B active.
- Judge models by task fit, licensing, context, and serving needs, not total parameter count alone.
- Test Inkling Small on real workloads before assuming its benchmark proximity translates to production reliability.
- Watch active MoE parameters closely because they shape practical deployment tradeoffs.