Time to Bail · Boarding pass
Act. You can stay.
Both leave serverless on October 22, 2026. Switch models or move to a dedicated endpoint first.[1]
Act means: change a setting or plan before the date. You can stay.
Together AI drops Llama-3.3-70B-Instruct-Turbo on October 22, 2026
Together AI removes two models from its serverless API on October 22, 2026: Llama 3.3 70B Turbo and DeepSeek V4 Flash 0731. Both can still run on a dedicated endpoint.
One email with the first steps. Nothing else.
Or add the date to your calendar: Apple or Outlook (.ics) · Google Calendar
Tick steps as you go. Your progress stays in this browser.
Search your code for Llama-3.3-70B-Instruct-Turbo and DeepSeek-V4-Flash-0731.
Together points to Muse-Glimmer-30B and DeepSeek-V4.1-Flash. Or keep the old model on your own endpoint.
Try the new model on a small share of calls first. Then move the rest. Watch errors and cost.
One email with the first steps. Nothing else.
Or add the date to your calendar: Apple or Outlook (.ics) · Google Calendar
| Removal date | October 22, 2026[1] |
|---|---|
| Llama model | meta-llama/Llama-3.3-70B-Instruct-Turbo → meta-models/Muse-Glimmer-30B[1] |
| DeepSeek model | deepseek-ai/DeepSeek-V4-Flash-0731 → deepseek-ai/DeepSeek-V4.1-Flash[1] |
| Short names | Llama 3.3 70B Turbo and DeepSeek V4 Flash 0731[1] |
| On-demand dedicated endpoints | Supported for both models[1] |
| Your options | An on-demand dedicated endpoint, a monthly reserved endpoint, or a newer serverless model[1] |
| Notice for serverless models | 2 or 3 weeks[1] |
| After the removal date | The model is no longer available through its serverless endpoint[1] |
| Apps that call these models on serverless | Stop working on October 22, 2026[1] |
Serverless calls to these two model IDs stop working. Together says the model is no longer on its serverless endpoint.
meta-models/Muse-Glimmer-30B or deepseek-ai/DeepSeek-V4.1-Flash.Source: [1] Deprecations (Together AI)
Check it yourself. Every fact above links here.
Time to bail on Together AI serverless: Llama 3.3 70B and DeepSeek? Act. You can stay. Both leave serverless on October 22, 2026. Switch models or move to a dedicated endpoint first.
On our scale this is Act: change a setting or plan before the date. You can stay.
Together AI removes two models from its serverless API on October 22, 2026: Llama 3.3 70B Turbo and DeepSeek V4 Flash 0731. Both can still run on a dedicated endpoint.
Here are the facts we checked:
You can stay on Together AI. You have to change a setting or plan before the date.
Leaving is a choice here, not a must.
Together AI is the main source for this page. Together AI has not said more than what is listed here. If a fact is not on this page, we do not know it yet.
Apps that call these models on serverless: Stop working on October 22, 2026.
The date to remember is October 22, 2026. Make the change before then.
A few tips:
If the change affects your members or team, send one short email. Say what changes for them and when. Put the most important line first.
If there is a date, send a short reminder a few days before it. Keep a copy of the list you sent it to.
Listed for convenience. Not ranked. Not paid. These names come up in threads and reviews; that does not show where people moved.
Before you pay, check the price page of each tool yourself. Prices change, and we only list what our sources say.
No questions collected yet.
Thanks.
Run SaaS spend at work? Get these changes before the invoice
Arrivals · for builders
People on Together AI serverless: Llama 3.3 70B and DeepSeek can stay if they make the change. Some would rather leave. Get the playbook for this change, and list your product here.
Listing is free. An editor checks it. Not ranked. Not paid.
One email when we publish a checked change in the category you pick. No account needed.
One email on Mondays: the gates in the next 14 days, with the first step for each.