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

Three new LLMs are now available for Telnyx Voice AI Assistants: GLM-5.2 from Z.ai, plus GPT-5.6 Luna and GPT-5.6 Sol from OpenAI. GLM-5.2 is the highest-intelligence open-weight model available, GPT-5.6 Luna offers a strong cost-to-intelligence ratio for voice workloads, and GPT-5.6 Sol rounds out the lineup as OpenAI's most capable model for complex conversations.

What's new

• zai-org/GLM-5.2: The highest-ranked open-weight model on Artificial Analysis with an Intelligence Index of 51, now selectable as the LLM for any Voice AI Assistant. 1M token context window for maintaining full conversation history without summarization.
• openai/gpt-5.6-luna: A cost-efficient closed-source option that trades some intelligence for lower latency and pricing. A practical combination of cost, latency, and intelligence for voice agents that handle high call volumes.
• openai/gpt-5.6-sol: OpenAI's most capable model, available for teams that need the highest quality responses on complex multi-step calls. Higher per-token cost makes it better suited for low-volume, high-value workloads than high-throughput voice campaigns.
• Mission Control selection: All three models are selectable from the model dropdown in the Assistant configuration, with no separate API keys or vendor contracts required.

Why it matters

• The cost gap between open-weight and closed-source models is narrowing from both directions. GLM-5.2 matches or beats frontier closed-source models on reasoning benchmarks at a fraction of the inference cost. GPT-5.6 Luna demonstrates that capable closed-source models can also compete on price, giving developers a broader choice than the old binary of cheap-and-limited vs smart-and-expensive.
• Running GLM-5.2 on Telnyx-hosted GPUs keeps the full voice pipeline on one private backbone. Audio enters the network, gets transcribed, routed to the LLM, synthesized back to speech, and delivered to the caller without leaving Telnyx infrastructure. Every vendor boundary that gets removed eliminates 30 to 80ms of latency.
• Model choice in Mission Control means teams can A/B test models on real call traffic, compare latency and quality side by side, and route different assistants to different models based on workload. A high-volume intake assistant can run on Luna for cost efficiency while a complex troubleshooting agent runs on GLM-5.2 for higher quality responses.

Example use cases

• High-volume outbound campaigns where per-call cost matters more than deep reasoning, using GPT-5.6 Luna to keep costs down at scale.
• Complex support flows that require multi-step handling and large context, using GLM-5.2 with its 1M token window to maintain full conversation history without summarization.
• Specialized assistants handling legal, financial, or technical queries where response quality matters more than cost, using GPT-5.6 Sol for low-volume, high-stakes calls.
• Teams migrating from Vapi or Retell who want to compare model performance on real telephony traffic before committing to one provider.

Getting started

1. In Mission Control, navigate to AI, then Assistants.
2. Create a new assistant or select an existing one.
3. In the model dropdown, select zai-org/GLM-5.2, openai/gpt-5.6-luna, or openai/gpt-5.6-sol.
4. Save and test your assistant with a trial call.

Learn more in the AI Assistants documentation or the GLM-5.2 release notes.