Publishers › Inception Labs
Inception Labs
Builds diffusion language models — the Mercury family — which generate tokens by iterative refinement of a whole sequence rather than strictly left to right. That makes their decode behaviour unlike the autoregressive models most of this registry describes, and the per-token throughput figures here are not directly comparable.
Who they are
Inception Labs is a frontier lab based in the United States, documented at www.inceptionlabs.ai. That makes their decode behaviour unlike the autoregressive models most of this registry describes, and the per-token throughput figures here are not directly comparable.
How they are funded
No ownership or funding information has been recorded for this organisation.
What they publish here
loading from the registry…