Skip to main content
Open-weight LLMs hosted on Telnyx GPU infrastructure. All chat models are accessible via the Chat Completions API (OpenAI-compatible); embedding models are served on the embeddings endpoints. For supported Flex, Default, and Priority options, see Service tiers. For current total-input, uncached-input, and output tokens-per-minute limits by model, see Inference token rate limits. Not every model here is available to AI Assistants, and of the assistant models only a subset is verified for voice calls — see Voice AI models.

​
Chat Models

Reasoning (thinking) behavior differs by surface: reasoning models return their chain-of-thought in a separate reasoning_content field on chat completions (see getting started), but reasoning is always disabled on voice calls and cannot be enabled there — see Reasoning on voice calls.

​
Embedding Models

All three models are available on the OpenAI-compatible Create embeddings endpoint (POST /v2/ai/openai/embeddings); List embedding models returns the current list programmatically. When you embed a Telnyx Storage bucket for AI use — via the Embed documents endpoint or the portal’s Embed for AI Use button (see Embeddings) — the embedding_model field accepts thenlper/gte-large and intfloat/multilingual-e5-large only, and defaults to intfloat/multilingual-e5-large. Qwen/Qwen3-Embedding-8B is available on the OpenAI-compatible embeddings endpoint only. Inputs longer than the model’s context length are rejected with a 400 error on the OpenAI-compatible endpoint.