Thinking Machines releases Inkling-Small, matching Inkling at quarter size
Inkling-Small is a Mixture-of-Experts transformer with 276B total parameters and 12B active, trained on NVIDIA GB300 NVL72 systems. It scores 31.6% on Humanity's Last Exam, ahead of Inkling's 29.7%, with a 1M-token context window. Full weights are available, plus fine-tuning on Tinker.
1 source
Daily brief
Get tomorrow's AI brief in your inbox
More stories today
- ArcKit generates architecture diagrams via AI-assisted workflow
- Guide to building AI agents with LangChain and Groq free tiers
- DeepAnalyze: agentic LLM auto-runs data science pipeline
- Reddit user shares AI video edit with I2VID workflow
- Tool gives AI agents persistent cross-session memory, zero API keys