# Model Serving - Scheduled for Retirement

The models listed here have been scheduled for decommissioning. Phoeniqs provides advance notice approximately 20 business days before the decommissioning date by publishing an announcement in Cloud Docs and notifying impacted customers directly. Communications include the expected timeline, recommended replacement models, and migration guidance where applicable, and are included in the monthly AI Platform newsletter shared with opt-in customers.

Access is not automatic

At 10 business days before the decommissioning date, the model is taken offline and API calls will return an error. If you still require access during this window, you can request temporary reactivation by opening a service ticket referencing the Model Name and the date you need access until. Without an open request, the model is permanently decommissioned on the stated date.

# Scheduled for Retirement

Model Name	Decommissioning Date	Model	Type	Input (Credits/M Tokens)	Output (Credits/M Tokens)	Suggested Replacement	Description
`inference-apertus-8b`	11.05.2026	swiss-ai/Apertus-8B-Instruct-2509	Chat	0,1324	0,1431	`inference-apertus-70b`	Optimized for multilingual dialogue use cases.
`inference-deepseekr1-70b`	11.05.2026	RedHatAI/DeepSeek-R1-Distill-Llama-70B-quantized.w4a16	Chat	0,4617	0,4617	`inference-qwq-32b`	Optimized for Reasoning chat completions.
`inference-deepseekr1-670b`	11.05.2026	RedHatAI/DeepSeek-R1-0528-quantized.w4a16	Chat	1,96	4,57	`inference-deepseek-v32`	Optimized for Reasoning chat completions.
`inference-kimi-k2`	11.05.2026	RedHatAI/Kimi-K2-Instruct-quantized.w4a16	Chat only	0,77	2,31	`kimi-K2.6`	Optimized for multilingual dialogue use cases.
`inference-llama33-70b`	11.05.2026	RedHatAI/Llama-3.3-70B-Instruct-quantized.w4a16	Chat only	0,5464	0,5464	`inference-llama4-maverick`	Optimized for multilingual dialogue use cases.
`inference-qwq25-vl-72b`	11.05.2026	RedHatAI/Qwen2.5-VL-72B-Instruct-quantized.w4a16	Multimodal	0,8465	0,8465	`inference-qwen3-vl-235b`	Optimized for compact and efficient vision-language model.

NOTE Once a model's decommissioning date has passed, it is permanently removed from the service and cannot be restored.

maas ai model serving scheduled-for-retirement