CorX Labs is an independent AI research lab in Jamaica, and the lab that publishes this index. It has released 3 models, all with open weights: a 27B Jamaican Patois assistant, a singing voice synthesis model, and a small language model trained from random weights. All 3 are listed below.
Released
Each one, what it is and what it is built from. Full model cards live in Our Products.
Jamaica's first large open-weight LLM — a Jamaican Patois-speaking assistant fine-tuned from Qwen3.8-27B and fully merged.
A 158M-parameter decoder-only transformer written and trained from scratch on a single GPU — architecture, tokenizer, pretraining and fine-tuning.
A singing voice synthesis model that builds the source-filter theory of the human voice into its architecture as a hard constraint — pitch, timbre and breath texture run as three parallel encoder streams, combined only in a late fusion trunk. Generates at 24 kHz through rectified flow matching in roughly 32 sampling steps.
Benchmarks
Every other lab in this index is listed with the numbers it published. CorX Labs has not run these evaluations on its own models, so it has no numbers to list — and it is not going to quote Qwen3.8-27B's scores as CorX3.8-27B's, estimate from model size, or put a figure in a column it did not measure. An index that made an exception for the lab that runs it would be worth nothing.
The specification rows are real and comparable, so the models can still be put side by side with anything else here on parameters, context, modalities, licence and cost. When the evaluations are run, the scores go in like everyone else's.
In the index
The same row every other model gets, so nothing about ours is presented differently.
| CorX3.8-27BCorX Labs | CorX Labs | 33K | — | — | — | — | — | — | Apache 2.0 |
| CorX1.5CorX Labs | CorX Labs | 1K | — | — | — | — | — | — | Apache 2.0 |
| TriStream-SVSCorX LabsSinging voice synthesis | CorX Labs | — | — | — | — | — | — | — | Apache 2.0 |
