Model Deprecation

TensorX keeps a small, current line-up of open-weight models. This page explains how models are retired, how you are told, and what happens to requests that still use a retired model ID.

Why we keep the line-up short

TensorX focuses on the latest open-weight models from the leading labs, served on NVIDIA B300 GPUs in our own EU data centres. We deliberately keep a short line-up: the current version of each model family and the one before it. That keeps capacity on the models most customers use, so they stay fast and available. Older model IDs are rerouted to their replacement rather than switched off, so existing integrations keep working.

Policy

We carry the current version of a model family and the one before it. When a new version is released, the oldest of the three is retired two weeks later.

Example: GLM-5.3 is released. GLM-5.2 stays available. GLM-5.1 is retired.

Every retirement names a replacement. The replacement is the current model in the same family and tier.

Every retirement is announced before it takes effect. Notice is sent by email to account holders and recorded in the table below.

Retired model IDs are rerouted, not switched off. A request that uses a retired ID is answered by its replacement, so existing workloads keep responding without a code change. You are only moved once: a retired ID always points straight at the current replacement.

Billing follows the model that serves the request. Once an ID is rerouted, requests are billed at the replacement model's rate. Current rates are on tensorx.ai/models.

Update your model IDs

Rerouting is a safety net, not a destination. Call the model you actually want so that you control which model runs, what it costs and what appears in your logs.

Old IDs will not resolve indefinitely. We will give notice before a rerouted ID stops resolving.

To check which IDs your key can use:

curl -sS https://api.tensorx.ai/v1/models \
  -H "Authorization: Bearer YOUR_API_KEY" \
  | jq -r '.data[].id'

See Models for capability and context-limit metadata.

Retired models

Retired model ID	Rerouted to	Notice	End date
z-ai/glm-5	z-ai/glm-5.3	23 September 2026	4 October 2026
z-ai/glm-5.1	z-ai/glm-5.3	23 September 2026	4 October 2026
z-ai/glm-5-turbo	z-ai/glm-5.3-flash	23 September 2026	4 October 2026
z-ai/glm-5v-turbo	z-ai/glm-5.3-flash	23 September 2026	4 October 2026
deepseek/deepseek-r1-0528	deepseek/deepseek-v4-pro-0813	23 September 2026	4 October 2026
deepseek/deepseek-v3.2	deepseek/deepseek-v4-pro-0813	23 September 2026	4 October 2026
deepseek/deepseek-v4-pro	deepseek/deepseek-v4-pro-0813	23 September 2026	4 October 2026
moonshotai/kimi-k2.5	moonshotai/kimi-k3	23 September 2026	4 October 2026
moonshotai/kimi-k2.6	moonshotai/kimi-k3	23 September 2026	4 October 2026
qwen/qwen3-235b-a22b-2507	qwen/qwen3.8-2.4t-a95b	23 September 2026	4 October 2026
qwen/qwen3.5-122b-a10b	qwen/qwen3.8-flash-next	23 September 2026	4 October 2026
qwen/qwen3.5-9b	qwen/qwen3.8-flash-next	23 September 2026	4 October 2026
minimax/minimax-m2.5	minimax/minimax-m3	23 September 2026	4 October 2026

The 4 October 2026 round cleared out older models in one go. Later retirements follow the two-week cadence above and are added to this table.

Questions

Any questions, contact support.