AI · 1h ago
Inkling MoE Model Launches with Token-Efficient Reasoning and Agent Safety Tools
Inkling, a 1T-parameter mixture-of-experts model with 40B active parameters, offers native multimodal I/O and a reasoning_effort API for per-request compute control. It is available on Together Serverless and Vercel's AI Gateway, while Microsoft and Vercel also released primitives for safer agent code execution. The model's hardware requirements are steep, with full precision needing 600GB VRAM.
Meridian48 take
The reasoning_effort parameter is promising but underdocumented, and the hardware bar limits self-hosting to well-resourced teams.
Read the full reporting
Inkling MoE + Agent Safety: Token Efficiency Meets Reliability →
DEV Community
mixture-of-expertsagent-safety