SunsetDiff

Together AI model shutdowns on November 6, 2026

The dataset has 2 Together AI ids with shutdown date 2026-11-06: MiniMaxAI/MiniMax-M3, thinkingmachines/Inkling. The deprecations page says these models will be removed from serverless inference on the date listed. Each row stores the recommended replacement from that table. The scheduled table marks these rows Yes for on-demand dedicated endpoints.

Last reviewed . Publisher: SunsetDiff.

Steps

  1. Read the ids and the replacement. The table is every Together AI id in the dataset whose shutdown date is 2026-11-06. The replacement cell is the recommended replacement stored from the scheduled deprecations table. Open the model page before you change a call.
  2. Check on-demand dedicated endpoints. The scheduled table marks these rows Yes for on-demand dedicated endpoints. The deprecations page says: "Models marked "Yes" can be deployed as on-demand dedicated endpoints, either under the listed ID or as the underlying base model of a serving variant (for example, a deprecated -FP8 or -Turbo ID)." A serverless call is the call the removal date ends. A dedicated endpoint is a separate deployment.
  3. Change the model field. The quickstart posts to https://api.together.ai/v1/chat/completions and sets an OpenAI client base_url of https://api.together.ai/v1. Put the replacement id from the same row in the model field. The quickstart sample already uses MiniMaxAI/MiniMax-M3.1-Flash.
  4. Test before the date. Call the replacement with the prompts you send now. Do that before November 6, 2026. The deprecations page says replacement models rarely behave identically, so validate quality, latency, and cost before committing.

Models shutting down

These rows are every Together AI model in the dataset with shutdown date 2026-11-06. The model id links to its SunsetDiff model page. Deprecation date, shutdown date, and replacement are the fields stored on that row. Not published means the field is empty. These rows do not store a deprecation_date.

The deprecations page says: "The following models are deprecated and will be removed from serverless inference on the date listed. Migrate to the recommended replacement or a dedicated endpoint before that date."

The recommended replacement is the official replacement stored on each row. MiniMaxAI/MiniMax-M3: MiniMaxAI/MiniMax-M3.1-Flash. thinkingmachines/Inkling: meta-models/Muse-Spark-Open-816B-A42B. The scheduled table marks these rows Yes for on-demand dedicated endpoints.

Qwen/Qwen3.8-2.4T-A95B is on the same scheduled table with removal date 2026-11-04. The recommended replacement cell is Qwen/Qwen3.7-Max. On-demand dedicated endpoints are marked No for that row. The stored status is retired.

2 Together AI model ids with shutdown date 2026-11-06.
Model idDeprecation dateShutdown dateRecommended replacement
MiniMaxAI/MiniMax-M3Not published2026-11-06MiniMaxAI/MiniMax-M3.1-Flash
thinkingmachines/InklingNot published2026-11-06meta-models/Muse-Spark-Open-816B-A42B

Source: Together AI deprecations , checked November 5, 2026.

Serverless removal and dedicated endpoints

On November 6, 2026 these ids reach the shutdown date stored in the dataset. The deprecations page says: "After the removal date, the model is no longer available via its serverless endpoint, but migration options are described below."

The same page says: "Models marked "Yes" can be deployed as on-demand dedicated endpoints, either under the listed ID or as the underlying base model of a serving variant (for example, a deprecated -FP8 or -Turbo ID)." The scheduled table marks these rows Yes for on-demand dedicated endpoints. An on-demand dedicated endpoint is one of the migration options on that page.

The active redirects table is a different list. That section says: "Requests to the original model ID are automatically routed to the upgraded version." MiniMaxAI/MiniMax-M3 and thinkingmachines/Inkling are in the scheduled deprecations table. Update the model field to the recommended replacement, or deploy a dedicated endpoint, before November 6, 2026.

A November row outside this table is marked No. The page says: "Models marked "No" are not available as on-demand endpoints and require migration to a different model or a monthly reserved dedicated endpoint."

Sources: Together AI deprecations , Together AI dedicated endpoints , checked November 5, 2026.

MiniMax's own API is a different provider

SunsetDiff also stores MiniMax API rows. The guide is MiniMax M2.5 and M2.7 shutdown on November 18, 2026. Those ids are MiniMax-M2.5, MiniMax-M2.5-highspeed, MiniMax-M2.7, MiniMax-M2.7-highspeed, provider minimax, shutdown date 2026-11-18. The provider page is MiniMax.

Together's MiniMaxAI/MiniMax-M3 row is provider together. Its stored replacement is MiniMaxAI/MiniMax-M3.1-Flash. Calls to that id use the Together host on this page.

Source: Together AI deprecations , checked November 5, 2026.

Find the model ids in your code

Search the repository for the ids in the table, then for the replacement ids. A hit is a candidate, not proof the call still runs. How to find every model ID in your codebase covers SDK argument names. The scan page builds a search from dataset ids that have a shutdown date.

The quickstart says: "If you're already using the OpenAI SDK, you can point it at Together's base URL (https://api.together.ai/v1) and keep the rest of your code the same." Its curl posts to https://api.together.ai/v1/chat/completions.

ripgrep

rg -n -w -e 'MiniMaxAI/MiniMax-M3' -e 'thinkingmachines/Inkling'

Sources: Together AI deprecations , Together AI quickstart , checked November 5, 2026.

Test the replacement before the cutoff

The deprecations page says to run a small A/B test comparing the replacement against your current model on representative traffic. It says replacement models rarely behave identically, so validate quality, latency, and cost before committing. Do that before November 6, 2026.

The quickstart's curl and OpenAI client set model to MiniMaxAI/MiniMax-M3.1-Flash. That is the stored replacement for MiniMaxAI/MiniMax-M3. The quickstart also says: "This is especially helpful with a reasoning model like MiniMax M3.1 Flash, which works through a problem before answering and can produce a lot of output." And: "A reasoning model's response has two parts: the step-by-step thinking and the final answer. The thinking arrives in reasoning_content or reasoning, depending on the model, and the answer arrives in content."

thinkingmachines/Inkling stores meta-models/Muse-Spark-Open-816B-A42B. Put that id in the same model field. The quickstart sample does not print the Inkling id.

The pre-migration test checklist is the longer list: evals, rejected fields, token limits, tools, price, and latency.

OpenAI-compatible client, model field set to MiniMaxAI/MiniMax-M3.1-Flash

from openai import OpenAI
client = OpenAI(base_url="https://api.together.ai/v1")
response = client.chat.completions.create(
    model="MiniMaxAI/MiniMax-M3.1-Flash",
    messages=[{"role": "user", "content": "Hello"}],
)

OpenAI-compatible client, model field set to meta-models/Muse-Spark-Open-816B-A42B

from openai import OpenAI
client = OpenAI(base_url="https://api.together.ai/v1")
response = client.chat.completions.create(
    model="meta-models/Muse-Spark-Open-816B-A42B",
    messages=[{"role": "user", "content": "Hello"}],
)

Sources: Together AI deprecations , Together AI quickstart , checked November 5, 2026.

Related SunsetDiff pages

All guides

Questions

Which Together AI ids shut down on November 6, 2026?

The dataset lists MiniMaxAI/MiniMax-M3, thinkingmachines/Inkling with shutdown date 2026-11-06. The deprecations page says: "The following models are deprecated and will be removed from serverless inference on the date listed. Migrate to the recommended replacement or a dedicated endpoint before that date."

Do on-demand dedicated endpoints stop on that date?

The deprecations page says: "After the removal date, the model is no longer available via its serverless endpoint, but migration options are described below." The scheduled table marks these rows Yes for on-demand dedicated endpoints. The page says: "Models marked "Yes" can be deployed as on-demand dedicated endpoints, either under the listed ID or as the underlying base model of a serving variant (for example, a deprecated -FP8 or -Turbo ID)." The scheduled table tells you to migrate to the recommended replacement or a dedicated endpoint before that date.

Is this the MiniMax API retirement?

This page is provider together. The MiniMax API guide covers MiniMax-M2.5, MiniMax-M2.5-highspeed, MiniMax-M2.7, MiniMax-M2.7-highspeed on provider minimax, shutdown date 2026-11-18. Together's MiniMaxAI/MiniMax-M3 replacement is MiniMaxAI/MiniMax-M3.1-Flash.

Which host does the call use?

The quickstart says: "If you're already using the OpenAI SDK, you can point it at Together's base URL (https://api.together.ai/v1) and keep the rest of your code the same." It posts to https://api.together.ai/v1/chat/completions. The sample model on that page is MiniMaxAI/MiniMax-M3.1-Flash.

Why is Qwen/Qwen3.8-2.4T-A95B missing from the table?

Qwen/Qwen3.8-2.4T-A95B stores shutdown date 2026-11-04. This table is shutdown date 2026-11-06. The scheduled table lists Qwen/Qwen3.8-2.4T-A95B with recommended replacement Qwen/Qwen3.7-Max.

More guides

Deprecation alerts

We email you before models shut down. Unsubscribe anytime.

Deprecation alert signup
Providers (optional)