Thinking Machines beats frontier models by 29.8% fewer errors using Bridgewater expert judgments

Launched Fugu, a multi-LLM orchestrator that matches Anthropic's top models on benchmarks

Unveiled native multimodal interaction model that simultaneously listens, sees, speaks, and uses tools

7B Conductor achieved SOTA on GPQA-Diamond and LiveCodeBench; paper accepted at ICLR 2026.

Announced Shinka Evolve system for autonomous AI architecture design

Entered strategic partnership with NVIDIA for Vera Rubin AI systems deployment

Formed landmark partnership with NVIDIA for frontier model training and customizable AI platforms

Introduced Doc-to-LoRA hypernetwork for efficient long-context processing

The AI Scientist published in Nature 2026.