So
Thinking Machines, which is part of the "spun-off" start-ups from OpenAI (Anthropic can be considered the first), released some time back their open-weight Inkling pair of models. With
Inkling totalling 975B parameters (41B active) and its little brother
Inkling-Small 276B (12B active), both are Mixture-of-Experts models and both represent what most can be said to represent the "West's" big answer to the Chinese dominance of bit open-weight models, such as DeepSeek v4 Pro, Kimi K3 or Qwen 3.8 all of which surpass the 1T parameter size.