LoginSign Up
← Back to all sparks
T

Telnyx

COMMS
Velocity7.5

Programmable voice, messaging, and connectivity APIs for developers.

telnyx.com ↗

Telnyx pulls the whole agent stack onto its own platform — and cuts voice AI loose from the phone line

voice aiagent runtimeedge computechannel consolidationmodel integrations
◆Current state
Telnyx is shipping on two tracks at once. The steady one is model plumbing: GPT-5.4-mini, Kimi K3 on Inference, then GLM-5.2 and two GPT-5.6 variants, each slotted into Voice AI Assistants at a different cost-versus-reasoning point. The consequential one is structural — Edge Compute launched an Agent Runtime with Functions, KV, Cloud Storage, and Inference all GA, and Voice AI conversations now run over a direct WebSocket with no telephony involved. An Email API also entered beta, adding another channel to the same bill.
◆Where it's heading
The company built its position on owning telephony infrastructure, and it is now spending that position rather than defending it. Letting assistants stream PCM16 audio straight from a web or mobile client means the voice product no longer requires a phone number, which opens it to developers who were never going to buy carrier services. Edge Compute is the other half: instead of renting Telnyx a channel and hosting the agent elsewhere, the agent's compute, state, and storage sit next to the media path. Channel breadth — SMS, voice, WhatsApp, now email — is being levelled under one API and one invoice.
◆Prediction
Expect StatefulActor and CloudFS to reach GA next, and the Email API to leave beta and get folded into the same unified billing pitch. The model-add cadence should continue at roughly its current pace as new releases appear.

◆Recent moves

  1. 13d ago

    GLM-5.2, GPT-5.6 Luna, and GPT-5.6 Sol Now Available for Telnyx Voice AI Assistants

    Three more LLMs land in Voice AI Assistants, spread across cost and reasoning tiers. Routine catalog maintenance — the pattern here is keeping the model menu current rather than changing what the product does.

    View source ↗
  2. 15d ago

    Telnyx Edge Compute Is Now Available

    ⚡ SPARK

    The most structural release in this window: Telnyx moves from carrying an agent's traffic to hosting the agent itself. Together with the WebSocket release days earlier, it marks the shift from telephony vendor to full agent platform.

    View source ↗
  3. 16d ago

    Voice AI Conversations Over WebSocket Now Available for Telnyx AI Assistants

    ⚡ SPARK

    Voice AI Assistants become reachable without a phone number, streaming PCM16 audio from any web or mobile client. It decouples the assistant product from the telephony business that has been Telnyx's foundation.

    View source ↗
  4. 20d ago

    Email API Now Available in Beta

    Email joins SMS, voice, and WhatsApp in beta, pitched as one API and one bill. A real widening of the channel surface, though it follows the well-worn CPaaS path of levelling channels rather than opening a new direction.

    View source ↗
  5. 22d ago

    Kimi K3 Now Available on Telnyx Inference

    A 2.8T-parameter open-source model with a 1M context window and native vision arrives on Telnyx Inference at $2.70 per million input tokens. Notable for the published price point, but it extends an existing catalog rather than redirecting it.

    View source ↗
  6. 27d ago

    GPT-5.4-mini Now Available for Telnyx Voice AI Assistants

    A compact, cost-efficient model option aimed at high-volume voice agent workloads. Part of the recurring cadence of tiering the assistant model menu by price and latency.

    View source ↗