AI research lab founded by ex-OpenAI CTO Mira Murati. Ships Tinker, a managed API for post-training: LoRA fine-tuning and RL over open-weight models (Llama, Qwen dense and MoE) via Python primitives, with distributed GPU infra abstracted away. Also researches full-duplex multimodal interaction models trained natively on audio, video, and text in 200ms micro-turns.
Thinking Machines just dropped the Inkling 2 Small weights on Hugging Face under Apache 2.0, opening up its 594B agent-first MoE for self-hosting and fine-tuning.
Thinking Machines Lab is offering up to $50,000 in Tinker credits to researchers tackling the hardest open-weight model safety problems, from tamper-resistant safeguards to worst-case risk forecasting.