LiveKit Inference: Please review deprecated Grok model mappings and pricing

Getting Started
,
1

It looks like LiveKit Inference’s xAI model mappings may need an update following xAI’s recent model retirement in Build voice, video, and physical AI | LiveKit.

xAI retired several Grok models on May 15, but the old model IDs continue to resolve by automatically routing requests to newer models. The important catch is that billing follows the model that actually serves the request, not necessarily the retired model you selected.

For example, LiveKit Inference still lists models such as:

  • Grok 4.1 Fast — $0.20 / 1M input, $0.50 / 1M output

  • Grok 4.1 Fast Reasoning — $0.20 / $0.50

  • Grok 4.20

  • Grok 4.20 Reasoning

  • Grok 4.20 Multi-Agent

If these are being transparently redirected by xAI, developers may unknowingly be charged at the newer model’s pricing.

It may be worth reviewing and updating the entire xAI model catalog in LiveKit Inference to ensure:

  • Available models accurately reflect xAI’s current offerings.

  • Displayed pricing matches the model that is actually billed.

  • Deprecated models are removed or clearly marked if they are transparently redirected.

Migration notice from xAI:

2

Hi Kaushal, Looking into this :eyes:

3

Thanks again for raising this.

I can confirm that devs were not being charged more.

The team will also make the necessary updates for pricing and model support, to align with Grok. These updates are still in flight but should appear soon.

4

Also it would be great if openai 5.6 series is added? especially luna as its faster ttft model even than 5.4 mini and has better intelligence as claimed by openai, so its very tempting to use for voice agents.