Bon, on peut charger le modèle IA et le faire tourner en local. Mais je cite opencode:
1. Get an API key at https://tinker.thinkingmachines.ai (API Keys) — set it as the TINKER_API_KEY env var or paste it via /connect.
2. Run /models in opencode and select Inkling.
Alternative (free tier): use NVIDIA NIM — run /connect, pick NVIDIA, add a key from build.nvidia.com, then select thinkingmachines/inkling.
Note: running the full 975B weights locally isn't practical (needs many GPUs); the hosted endpoints above are the standard way to use it.
Donc pas essayé perso ... mais ceux avec pleins de GPU pourront =)

Source: thinkingmachines.ai