Choose a model
Choose a model by measured workflow performance, not provider reputation or model name. The right model must understand the calls, follow the system prompt, use attached actions correctly, respond naturally, and fit the target credit cost.
Models available in Modular
OpenAI
| Model shown in Wirevox | Model ID |
|---|---|
| GPT-4o Mini (Default) | gpt-4o-mini |
| GPT-4o | gpt-4o |
| GPT-4.1 | gpt-4.1 |
| GPT-4.1 Mini | gpt-4.1-mini |
| GPT-4.1 Nano | gpt-4.1-nano |
| GPT-5 Mini | gpt-5-mini |
| GPT-5 Nano | gpt-5-nano |
| GPT-5.1 | gpt-5.1 |
| GPT-5.2 | gpt-5.2 |
| GPT-5.4 | gpt-5.4 |
| GPT-5.4 Mini | gpt-5.4-mini |
| GPT-5.4 Nano | gpt-5.4-nano |
GPT-4o Mini is the current Wirevox default for Modular agents. Treat Mini and Nano variants as candidates for focused, high-volume workflows, then verify that they follow every consequential branch and action rule.
Larger models are candidates when the workflow has more ambiguity, policies, or tool dependencies. A larger model does not correct an incomplete prompt, missing source of truth, or unsafe function design.
Grok
| Model shown in Wirevox | Model ID |
|---|---|
| Grok 4.20 Fast | grok-4.20-non-reasoning |
Benchmark Grok 4.20 Fast when you want to compare xAI prompt adherence and action behavior with the OpenAI and Gemini modular options. Selecting it does not give the agent automatic web or X search; search requires a capability supplied by Wirevox.
Google Gemini
| Model shown in Wirevox | Model ID |
|---|---|
| Gemini 1.5 Flash | gemini-1.5-flash |
| Gemini 2.0 Flash | gemini-2.0-flash |
| Gemini 2.5 Flash | gemini-2.5-flash |
Gemini 2.5 Flash is Wirevox's default within the Gemini modular provider. Compare the Flash options with the same prompt, voice, STT provider, and tests so the model is the primary changed variable.
Anthropic
Claude 3.5 Sonnet, Claude 4.5 Sonnet, and Claude 4.6 Sonnet can appear as Coming Soon. They are disabled and cannot currently be selected for a production Wirevox agent.
Models available in Realtime
OpenAI Realtime
| Model shown in Wirevox | Model ID |
|---|---|
| GPT Realtime (Existing) | gpt-realtime |
| GPT Realtime 1.5 | gpt-realtime-1.5 |
| GPT Realtime 2 | gpt-realtime-2 |
| GPT Realtime 2.1 (Latest) | gpt-realtime-2.1 |
| GPT Realtime 2.1 Mini | gpt-realtime-2.1-mini |
Use the version labels shown in the editor. Do not assume an older agent is automatically moved to the model labeled Latest.
Grok Realtime
| Model shown in Wirevox | Model ID |
|---|---|
| Grok Voice Think Fast 2.0 | grok-voice-think-fast-2.0 |
| Grok Voice Think Fast 1.0 | grok-voice-think-fast-1.0 |
Grok Voice Think Fast 2.0 is Wirevox's default within the Grok Realtime provider. The grok-voice-latest alias is registered internally but is not selectable in the editor, which avoids an unreviewed alias change in a saved configuration.
Gemini Live
| Model shown in Wirevox | Model ID |
|---|---|
| Gemini Live | gemini-live |
Gemini Live uses a compatible Gemini realtime voice and provider-managed speech session.
Use a decision sequence
1. Decide whether architecture requirements eliminate options
If the agent needs a secondary voice for another language, start with Modular. If it needs a provider-native speech-to-speech session, start with Realtime.
2. Establish a baseline
Use the current default or the existing published model and record:
- task completion;
- factual accuracy;
- function selection and arguments;
- confirmation behavior;
- interruption and silence behavior;
- time to the first useful response;
- average response length; and
- estimated credits per minute.
3. Build representative tests
Include ordinary requests, ambiguity, corrections, missing information, unsupported requests, action success, no-result and failure, escalation, noisy speech, silence, and interruption.
4. Compare one change at a time
Keep the prompt, functions, STT, voice, and conversation controls unchanged when comparing two Modular language models. For Realtime comparisons, recognize that changing provider also changes the compatible voice and native speech behavior.
5. Select the least expensive model that passes
Do not optimize only for the lowest displayed rate. A cheaper model that causes incorrect bookings, unnecessary transfers, longer calls, or repeat contacts can cost more operationally.
6. Retest after prompt changes
A model comparison is valid only for the prompt and tool set tested. Run the suite again after materially changing the system prompt, attached actions, knowledge, or architecture.
Read the cost estimate correctly
The voice-agent header displays the current draft's estimated credits / min and a component breakdown. Modular estimates can include platform, STT, LLM, and TTS components. Realtime uses the selected realtime pipeline rate plus applicable platform components.
The rate is configuration-dependent and can change as costs are updated. Use the live editor rather than a static documentation table for the current estimate. Separately priced function use can add charges beyond the per-minute voice rate.
Avoid these selection mistakes
- Choosing the model with the largest version number without testing the workflow.
- Comparing models while also changing the prompt, voice, and conversation settings.
- Assuming a text-model result predicts Realtime speech behavior.
- Assuming model knowledge is live business data.
- Depending on a Coming Soon or non-selectable model.
- Ignoring the published version while testing only the draft.
- Optimizing the per-minute estimate without measuring completion and call duration.