OpenAI Model Changes - Cost & Usage Guide
- Introduction
- Cost vs Accuracy Ratio
- Model Cost Tiers (Highest to Lowest)
- Focus on Affordability
- Focus on Accuracy
- Deprecating Models
- Why are they Being Deprecated?
- Transcription Models (PBX Transcription)
- Call Recording EndPoint APIs Used
- Models
- Text & Analysis Models (PBX Analytics)
- Summary/Sentiment EndPoint APIs Used
- Models
- Realtime Models (Live Calls & AI Agents)
- Calls Endpoint
- Models
- gpt-realtime-1.5
- Text Endpoint
- Models
Introduction
This guide explains the range of OpenAI models available, their associated costs and how each should be used for PBX calls, transcription, analytics, and AI agents. The main aims of this guide can be summarised as follows:
- Reduce operating costs
- Standardise defaults
- Make model selection predictable and explainable
For users who want to compare OpenAI models and pricing in more detail, OpenAI provides an official reference page: https://platform.openai.com/docs/models/
Cost vs Accuracy Ratio
Smaller models are more affordable, but less accurate.
- nano < mini < base < versioned < pro
This rule holds across all OpenAI model families.
Model Cost Tiers (Highest to Lowest)
Focus on Affordability
These models are safe for high‑volume PBX workloads such as call summaries, sentiment analysis, and CRM notes.
Model | Cost Level | Intended Use |
gpt-5.4-nano | Lowest | Extremely high volume, simple tasks |
gpt-5.4-mini | Low | Default model – best price/performance |
gpt-5.4-mini is set as the default as it offers the best quality-to-cost ratio.
Focus on Accuracy
These models should only be used when higher accuracy or reasoning is explicitly required.
Model | Cost Level | Notes |
gpt-5.4 | High | Advanced summaries, QA audits |
Deprecating Models
OpenAI deprecates models as they invest in newer, more efficient architectures.
Reference: https://developers.openai.com/api/docs/deprecations
Why are they Being Deprecated?
- Older architecture
- Worse cost/performance than current mini
- Less affordable for the same output quality
Transcription Models (PBX Transcription)
Due to these changes, we now expose model selection under: System → Integrations → Transcription → OpenAI.
This allows Administrators to choose the transcription model that best fits their cost and quality requirements.
Call Recording EndPoint APIs Used
- v1/audio/transcriptions
- v1/audio/translations
Models
Model | Cost Level | Intended Use |
Whisper | Most Affordable | Default for bulk call recording |
gpt-4o-mini-transcribe | Medium | Better formatting |
gpt-4o-transcribe | Higher | Highest transcription quality |
Whisper remains the most cost‑effective option for PBX transcription.
Text & Analysis Models (PBX Analytics)
Summary/Sentiment EndPoint APIs Used
- v1/chat/completions
Models
- gpt-5.4 (Default)
- gpt-5.4-mini
- gpt-5.4-nano
Realtime Models (Live Calls & AI Agents)
You can use the “Pricing” links to evaluate the cost differences between the offered models.
Calls Endpoint
- v1/realtime
Models
gpt-realtime-1.5
Text Endpoint
- v1/responses
Models
- gpt-5.4-mini (Default)
- gpt-5.4 (High Accuracy)
Last Updated
This document was last updated 26 May 2026