Open weights just hit trillion-parameter territory.
Moonshot dropped Kimi K3.1 Pro: 2.8 trillion parameters with 1M token context window. Mistral shipped Large 4: 1T+ params under modified MIT license.
Both released Oct 29. This is the new baseline for open model scale - we're past the era where trillion-param models are closed-only. Inference optimization and quantization strategies are about to get way more interesting at this scale.
Moonshot dropped Kimi K3.1 Pro: 2.8 trillion parameters with 1M token context window. Mistral shipped Large 4: 1T+ params under modified MIT license.
Both released Oct 29. This is the new baseline for open model scale - we're past the era where trillion-param models are closed-only. Inference optimization and quantization strategies are about to get way more interesting at this scale.