Model lifecycle

Learn about Serverless Inference model lifecycle and retirement

This page describes how models in the Serverless Inference catalog move through their lifecycle stages, how W&B notifies you of upcoming changes, and which models are deprecated or retired. Use this information to plan migrations away from models that are scheduled for removal.

W&B evaluates and releases new models as part of the Serverless Inference catalog on an ongoing basis.

W&B reviews older models periodically and may retire them to prioritize the most in-demand and highest-quality offerings.

Model lifecycle stages

Models in Serverless Inference use the following terminology for the stages of their lifecycle:

• Experimental: The model is available for initial experimentation, but isn’t bound by our standard SLA or deprecation process.
• Generally available: The model is fully supported and recommended for use.
• Deprecated: The model is scheduled for removal. Update to a recommended replacement as soon as possible.
• Retired: The model is no longer available. Any requests to retired models fail and return an HTTP 404 status code.

Deprecation process

When W&B deprecates a generally available model, we provide at least two weeks of notice on this page before the model is retired.

We also attempt to notify recent users of a deprecated model through email and provide recommendations for replacement models.

Models that are in the deprecated stage continue to serve requests.

Deprecated models

The following Serverless Inference models are deprecated:

Deprecation date	Retirement date	Model ID	Recommended replacement
2026-06-30	2026-07-20	microsoft/Phi-4-mini-instruct	JetBrains/Mellum2-12B-A2.5B-Instruct
2026-06-30	2026-07-20	nvidia/Kimi-K2.5-NVFP4	moonshotai/Kimi-K2.6, moonshotai/Kimi-K2.7-Code
2026-06-30	2026-07-20	Qwen/Qwen3.5-27B	Qwen/Qwen3.6-27B
2026-06-30	2026-07-20	Qwen/Qwen3-235B-A22B-Instruct-2507	Qwen/Qwen3.6-27B
2026-06-30	2026-07-20	Qwen/Qwen3-235B-A22B-Thinking-2507	Qwen/Qwen3.6-27B

Retired models

The following Serverless Inference models are retired:

Retirement date	Model ID	Recommended replacement
2026-05-11	zai-org/GLM-5-FP8	zai-org/GLM-5.1
2026-04-21	meta-llama/Llama-4-Scout-17B-16E-Instruct	Qwen/Qwen3.5-35B-A3B
2026-03-04	deepseek-ai/DeepSeek-R1-0528
2026-03-04	deepseek-ai/DeepSeek-V3-0324
2026-03-04	moonshotai/Kimi-K2-Instruct	moonshotai/Kimi-K2.5
2026-03-04	moonshotai/Kimi-K2-Instruct-0905	moonshotai/Kimi-K2.5
2026-03-04	Qwen/Qwen2.5-14B-Instruct	OpenPipe/Qwen3-14B-Instruct
2026-03-04	zai-org/GLM-4.5	zai-org/GLM-5-FP8