Skip to main content
On December 15, 2025, fixie-ai/ultravox-qwen3-32b-preview is being removed. It will be replaced with GLM 4.6 as the recommended model for most use cases.

​
Why Change?

We’re removing Qwen3 32B due to low usage and because GLM 4.6 outperforms it across all our benchmarks. GLM 4.6 offers:
  • Better instruction following
  • More reliable tool calling
  • Improved accuracy with numbers and precise information
This change represents a significant upgrade for voice AI applications while maintaining the same pricing.

​
Timeline

  • December 3, 2025: Deprecation announced
  • December 15, 2025: fixie-ai/ultravox-qwen3-32b-preview will no longer work

​
How to Migrate

Update your model string to use GLM 4.6: 
# Before
model = "fixie-ai/ultravox-qwen3-32b-preview"

# After
model = "fixie-ai/ultravox-v0.7"
Testing Your Prompts: GLM 4.6 is a better instruction follower than Qwen3. If you’ve emphasized importance with phrases like “you must…” or “always…”, ensure these don’t harm performance with the new model. GLM 4.6 takes instructions seriously and may over-index on strongly worded directives.

​
Option 2: Alternative Models

If your use case has specific requirements, you can choose from other available models: Llama 3.3 70B: 
model = "fixie-ai/ultravox-llama3.3-70b"
Gemma 27B: 
model = "fixie-ai/ultravox-gemma3-27b-preview"
See the News page for a complete list of available models and their status.

​
What Happens If You Don’t Migrate?

Starting December 15, 2025, any call creation requests using fixie-ai/ultravox-qwen3-32b-preview will be rejected.

​
Pricing Impact

None. GLM 4.6 and all other models remain at $0.05 per minute.
Need Help?If you need assistance with this migration, please visit our Discord community.