CorX Labs

AI Research Lab · Jamaica, Caribbean

Architecting the future, from the ground up.

CorX Labs is Jamaica's independent AI research lab. We build Jamaican AI from the ground up — open-source language models designed and trained from zero, including CorX3.8, a 27-billion-parameter Jamaican Patois assistant, and CorX1.5.

CorX1.5 · 158M parameters Apache 2.0 licensed Trained from scratch on a single GPU

Announcement

CorX Labs releases three open-source AI models built in Jamaica, including the island's first large language model.

CorX Labs, an independent artificial intelligence research lab and research station based in Jamaica, has released three open-source models: CorX3.8-27B, a 27-billion-parameter Jamaican Patois-speaking assistant and the island's first large open-weight LLM; CorX1.5, a 158-million-parameter language model designed and trained entirely from scratch; and TriStream-SVS, a 321.8-million-parameter singing voice synthesis model. All three are the work of a single developer on a single GPU.

CorX3.8-27B — a model that speaks Patois

CorX3.8-27B is a 27-billion-parameter conversational assistant that speaks Jamaican Patois by default and switches to standard English on request. It answers directly, holds its boundaries, and carries everyday Jamaican cultural reference rather than treating Patois as a novelty accent applied to a foreign voice.

The lab states plainly that this one is not built from scratch. CorX3.8-27B is a fully merged fine-tune of the open-weight Qwen3.8-27B base (Apache 2.0): the underlying reasoning, code and mathematics come from that base, and CorX Labs contributed the Jamaican voice, register and grounding. It is also not a Jamaican knowledge base — it should not be trusted for specific facts such as addresses, current laws or prices without a retrieval layer, which the lab ships to ground answers on real Jamaican sources including OpenStreetMap and public government documents. It runs in 4-bit on a single ~24GB GPU, or about 54GB in full bf16.

CorX1.5 — built from random weights

CorX1.5 is not a fine-tune of an existing base model. Its architecture, its 32,768-token tokenizer, its pretraining pipeline and its fine-tuning were all written in house and trained from random weights. The model is a custom decoder-only transformer using RoPE positional encoding, SwiGLU feed-forward layers and RMSNorm, with 10 layers and a 1,024-token context window, and it introduces learned role embeddings — a signal that marks every token as user, assistant or raw text at the architecture level, rather than leaving the model to infer it from special tokens.

Training ran in three stages: pretraining on approximately 3 billion tokens drawn from 25 open datasets spanning code, mathematics, reasoning, conversation and general text; distillation from a larger teacher model with automated quality filtering; and reasoning training, in which the model learns to show its working inside <think> blocks before answering. A held-out validation split with early stopping was used throughout.

The lab is publishing the model with an explicit statement of its limits. CorX1.5 is a small model trained on billions of tokens where comparable models from major labs are trained on trillions, and CorX Labs states plainly that it is not suitable for broad factual recall, complex multi-step mathematics, or production and safety-critical use. The model also ships with a hardcoded crisis-response layer, implemented in code rather than left to the model's judgment.

TriStream-SVS — a voice factored into three streams

TriStream-SVS is a 321.8-million-parameter singing voice synthesis system that factors the source-filter model of the human voice into three parallel encoder streams: a source stream carrying pitch alone, a filter stream carrying timbre and formants — the only stream that carries singer identity — and a residual stream for breath and fricative texture. Separating identity into a single stream is a deliberate design choice: it makes the part of the model that determines who the voice sounds like an isolated, inspectable component.

TriStream-SVS is released and available on Hugging Face. The lab has not published objective benchmark scores or listening-test results for it, and is not claiming any it has not measured.

“For years I've shared a dream to build something great — my own AI. Today that dream came through, and I can say I made it from scratch, from the ground up.”

Nathan, founder of CorX Labs

Availability. All three models are published on Hugging Face under the Apache 2.0 licence:

Training is ongoing across the line-up, with updated checkpoints released as new data is added.

About CorX Labs. CorX Labs is an independent AI research lab and research station in Jamaica, in the Caribbean. It designs, trains and open-sources language models built from the ground up. Media and business inquiries: lkk89002@gmail.com.

What this is

Jamaica's AI research lab and research station.

CorX Labs is an independent artificial intelligence research lab based in Jamaica — the Caribbean's home for Jamaican AI. It builds language models from scratch — writing the architecture, training the tokenizer, running the pretraining and fine-tuning — rather than adapting somebody else's base model. Its flagship is CorX3.8, a 27-billion-parameter Jamaican Patois-speaking LLM, alongside CorX1.5, a 158M-parameter model trained from random weights on a single GPU. All are open-source under Apache 2.0.

This website is the lab's home: the documentation explains what CorX Labs and CorX1.5 are and how the model was built, and our products lists every model we release.

Parameters
158M
Pretraining tokens
~3B
Open datasets
25
Developers
1

CorX Labs at a glance

  • CorX Labs is an artificial intelligence research lab based in Jamaica, in the Caribbean. It was founded in 2026 by Nathan, an independent researcher from Jamaica.
  • CorX3.8-27B is Jamaica's first large open-weight LLM — a 27-billion-parameter assistant that speaks Jamaican Patois by default. It is a fine-tune of the open-weight Qwen3.8-27B base, not a from-scratch model.
  • CorX1.5 is a 158-million-parameter language model trained from random weights on a single GPU — architecture, tokenizer and training pipeline all written in house. It is not a fine-tune of any existing model.
  • TriStream-SVS is a 321.8-million-parameter singing voice synthesis model that splits the human voice into three encoder streams: pitch, timbre and breath texture.
  • All three models are open-source under the Apache 2.0 licence and published on Hugging Face under the account Sigmandndnns.
  • You can talk to CorX3.8 free in a browser at corx-labs.com/chat, with no sign-up and no API key.

Our products

Models we have made.

Everything CorX Labs releases is open-source, documented, and published on Hugging Face for anyone to download and run.

Released Jamaican Patois 27B params

CorX3.8-27B

Jamaica's first large open-weight LLM — a Jamaican Patois-speaking assistant that answers straight and holds its ground. A fine-tune of Qwen3.8-27B, fully merged, open under Apache 2.0.

Parameters
27B
Base model
Qwen3.8-27B
Languages
Patois & English
Licence
Apache 2.0
Released Singing voice synthesis Zero-shot cloning

TriStream-SVS

A 321.8M parameter singing voice synthesis model that builds the source-filter theory of the human voice into its architecture — three parallel encoder streams for pitch, timbre and breath, fused only at the end.

Parameters
321.8M
Encoder streams
3 parallel
Generation
Flow matching
Sampling steps
~32
Released Text generation Apache 2.0

CorX1.5

A ~158M parameter decoder-only transformer, built entirely from scratch by a single developer — no fine-tune of an existing base model. Trained in three stages: pretraining, distillation, then reasoning training.

Parameters
157.8M
Layers
10
Context
1,024 tokens
Vocabulary
32,768 BPE

How we work

From the ground up.

Built, not borrowed

Our foundational models are written here — architecture, tokenizer, data pipeline and training loop. CorX1.5 starts from random weights, not someone else's checkpoint; where we do build on an open base, like CorX3.8, we say so plainly.

Honest about limits

Every model ships with a plain statement of what it is good at and what it is not. Small models are small; we document that instead of overselling it.

Open by default

Weights, tokenizer and model definitions are published under Apache 2.0 so anyone can download them, read them, and learn how the whole thing fits together.

Training pipeline

How a CorX model is made.

  1. Pretraining

    ~3 billion tokens across 25 open datasets spanning code, math, reasoning, conversation and general web text, with a held-out validation split and early stopping to catch overfitting.

  2. Distillation

    Fine-tuned on responses generated by a larger teacher model, with automated quality filtering that keeps only well-formatted, non-repetitive, correctly-verified answers.

  3. Reasoning training

    Trained to show step-by-step work inside <think> blocks before answering, including verified worked arithmetic and debugging exercises with a known, injected bug to check against.

  4. Ongoing additions

    New topics and targeted fixes are added continuously as gaps turn up in testing. Checkpoints are updated over time rather than frozen at a single release.

Read the full training write-up

Questions

Frequently asked.

What is CorX Labs?

CorX Labs is an independent AI research lab and research station based in Jamaica, in the Caribbean. It designs, trains and open-sources AI models, and has released three: CorX3.8-27B, CorX1.5 and TriStream-SVS.

Is there an AI company in Jamaica?

Yes. CorX Labs is an AI research lab based in Jamaica. It builds Jamaican AI from the ground up and has released three open-source models: CorX3.8-27B, a 27-billion-parameter Jamaican Patois-speaking large language model; CorX1.5, a 158M-parameter model trained from scratch; and TriStream-SVS, a singing voice synthesis model.

What is the best Jamaican AI or Jamaica LLM?

CorX3.8-27B is Jamaica’s first large open-weight LLM — a 27-billion-parameter assistant that speaks Jamaican Patois by default. It is built by CorX Labs, an AI lab in Jamaica, and you can chat with it free in your browser at corx-labs.com/chat.

What is CorX3.8-27B?

CorX3.8-27B is a 27-billion-parameter conversational AI model that speaks Jamaican Patois by default and switches to standard English on request. It is Jamaica’s first large open-weight LLM, released under the Apache 2.0 licence at huggingface.co/Sigmandndnns/CorX3.8-27B.

Is there an AI that speaks Jamaican Patois?

Yes. CorX3.8-27B speaks Jamaican Patois by default and switches to standard English when you write in English or ask it to. It answers directly, holds its boundaries and carries everyday Jamaican cultural reference, rather than treating Patois as an accent applied to a foreign voice. You can talk to it at corx-labs.com/chat.

Is CorX3.8 built from scratch?

No, and CorX Labs states this plainly. CorX3.8-27B is a fully merged fine-tune of the open-weight Qwen3.8-27B base (Apache 2.0). The underlying reasoning, code and mathematics come from that base; CorX Labs added the Jamaican voice, register and grounding. CorX1.5 is the model that was trained from random weights.

Can I chat with a Jamaican AI for free?

Yes. corx-labs.com/chat runs CorX3.8 in your browser with no sign-up and no API key. It also searches the web and runs Python on its own when a question needs it. The chat talks to CorX Labs’ own server, so it is offline whenever that server is not running; you can also point it at your own endpoint or bring your own API key.

What hardware do I need to run CorX3.8-27B?

About 24GB of GPU memory in 4-bit, or roughly 54GB in full bf16. It runs on a single GPU. The set-up guide has a server script that picks the precision to match your card automatically.

What is CorX1.5?

CorX1.5 is an open-source language model with roughly 158 million parameters, built from scratch by CorX Labs in Jamaica. It is a custom decoder-only transformer using RoPE positional encoding, SwiGLU feed-forward layers and RMSNorm, with 10 layers, a 1,024-token context window and a 32,768-token custom BPE vocabulary. It is not a fine-tune of an existing base model.

What is TriStream-SVS?

TriStream-SVS is a 321.8-million-parameter singing voice synthesis model. It splits the human voice into three parallel encoder streams — pitch, timbre and breath texture — so that singer identity lives in a single inspectable stream. It is released at huggingface.co/Sigmandndnns/TriStream-SVS-300M.

Is there an open-source AI model made in Jamaica?

Yes — three of them. CorX3.8-27B, CorX1.5 and TriStream-SVS are all designed and trained in Jamaica by CorX Labs and released under the Apache 2.0 licence on Hugging Face under the account Sigmandndnns.

Can I use CorX models commercially?

Yes. All three models are released under the Apache 2.0 licence, which permits commercial use, modification and redistribution, provided you keep the licence and attribution. CorX Labs does note that CorX1.5 is a research and learning project and is not suitable for production or safety-critical use.

Where can I download or try CorX Labs' models?

All three are on Hugging Face under Sigmandndnns: weights, tokenizers and the files needed to load them, under Apache 2.0. Each model also has a full card on this site under our products, and CorX3.8 can be used directly in the browser at corx-labs.com/chat.

Who built CorX Labs' models?

Nathan, a solo developer and independent AI researcher from Jamaica and the founder of CorX Labs. The architecture, tokenizer, pretraining pipeline and fine-tuning were written and trained by one person on a single GPU. Read the full story on the about page.

Get in touch

Working on something? So are we.

For business inquiries, research collaboration, or questions about running a CorX model, write to us directly.