Our own model layer, owned by TLearth.

TLearth One serves open-weights models on scale-to-zero GPUs behind one OpenAI-compatible API. TLearth.com and any third-party platform can use it with a single key.

Get an API key Read the docs

curl https://<your-domain>/api/public/v1/chat/completions \
  -H "Authorization: Bearer tlk_live_..." \
  -d '{"model":"tlearth-one-chat","messages":[{"role":"user","content":"Hello"}]}'

tlearth-one-chat

TLearth One Chat

General reasoning, writing, coding and tool use.

Qwen2.5-32B-Instruct (AWQ) · Apache-2.0 · 1x A100 80GB

tlearth-two-code

TLearth Two Code

Specialized agentic coding model: writes and fixes code, plans multi-step tasks and calls tools.

Qwen2.5-Coder-32B-Instruct (AWQ) · Apache-2.0 · 1x A100 80GB

tlearth-one-vision

TLearth One Vision

Image understanding, OCR and document analysis.

Qwen2.5-VL-7B-Instruct · Apache-2.0 · 1x L40S

tlearth-one-embed

TLearth One Embed

Private semantic search and knowledge bases.

BGE-M3 · MIT · 1x L4

tlearth-one-voice

TLearth One Voice

Speech input and spoken answers.

Whisper large-v3-turbo + Kokoro-82M · MIT / Apache-2.0 · 1x L4

tlearth-one-private

TLearth One Private

The same containers packaged for offline or dedicated deployment on your own hardware.

Where we actually are

Now

TLearth AI

Managed multi-model access. Requests are routed to the best available backend.

First owned release

TLearth One

TLearth-hosted, permissively licensed open weights on our own serverless GPUs, customized through private context and policy.

Later

TLearth-trained

A model continued-trained by TLearth, if and when we invest in our own data and compute program.