Essential AI releases Rnj-1 model weights

The month is established; the day conflicts

Essential AI published its Rnj-1 announcement in December, while the model card identifies a different initial-version day in the same month. Both establish the model family and public release, but they do not consistently establish one day. This event deliberately uses month precision. [1][2]

The base and instruction-tuned models have 8 billion parameters. The instruction checkpoint uses Apache 2.0 and provides local inference instructions. Essential AI's evaluations describe its own testing, rather than an independent guarantee of coding performance. [1][2]