ElevenLabs deprecated four OpenAI models for its ElevenAgents voice agents in its November 17, 2026 changelog. The models are `gpt-4.1`, `gpt-4.1-mini`, `gpt-4o` and `gpt-4o-mini`. They shut down on January 14, 2027, and requests start falling back to `gpt-6-luna` on December 15, 2026. We read the changelog on November 22, 2026. Every date below comes from that page.
Key Takeaways
- ElevenAgents deprecated `gpt-4.1`, `gpt-4.1-mini`, `gpt-4o` and `gpt-4o-mini` on November 17, 2026, ahead of a January 14, 2027 shutdown.
- Fallback to `gpt-6-luna` covers 10% of requests from December 15, 2026 and all requests from January 9, 2027.
- Agents on `gpt-6-luna` with no reasoning effort set now run at `low` instead of `medium`.
- ElevenLabs also cut Speech to Text prices to $0.20 per hour for batch and $0.22 per hour for realtime on self-serve and Business plans.
- Topcalls AI calls start from $0.25 a minute, depending on the country called, with transcription included in the AI price.
Which ElevenAgents models are deprecated?
Four OpenAI models: `gpt-4.1`, `gpt-4.1-mini`, `gpt-4o` and `gpt-4o-mini`. The changelog says they are deprecated ahead of their shutdown on January 14, 2027. An agent that still points at one of them keeps working for now. After the fallback dates, the platform answers with a different model.

The same entry covers older models. The deprecated `gpt-5`, `gpt-5-mini` and `gpt-5-nano` now fall back to `gpt-6.1-sol`, `gpt-6.1-terra` and `gpt-6-luna`, in that order. Before this change, the planned replacements were the GPT-5.6 models announced on November 2, per the changelog.
When does the fallback start?
Four dates matter, and all are in the ElevenLabs changelog for November 17, 2026.
| Date | What happens |
|---|---|
| November 17, 2026 | `gemini-3.6-flash` falls back to `gemini-3.8-flash` for all requests, ahead of its retirement on November 19 |
| December 15, 2026 | 10% of requests to the four deprecated GPT models fall back to `gpt-6-luna` |
| January 9, 2027 | All requests to those models fall back to `gpt-6-luna` |
| January 14, 2027 | The four GPT models shut down |
Models already past their provider shutdown dates resolve to their replacements immediately. The changelog names Claude 3.5 Sonnet, Claude 3.7 Sonnet, Claude 3 Haiku, Claude Sonnet 4, Gemini 2.0 Flash, Gemini 2.0 Flash-Lite, `gpt-4-turbo` and `gpt-3.5-turbo`.
Will a model fallback change how my voice agent sounds?
It can change what the agent says and how it reasons, because the model writes the replies. The changelog does not describe how `gpt-6-luna` answers differ from `gpt-4o`, so no one outside ElevenLabs can promise the same behavior. Test before the 10% window opens.
One behavior change is documented. Agents on `gpt-6-luna` that have no reasoning effort setting now run at `low` instead of `medium`. If your agent never set a value, its behavior moved without a change on your side.
What else shipped on November 17?
Four items touch voice agent teams, per the same changelog:
- Speech to Text prices: every Scribe model costs $0.20 per hour of audio for batch and $0.22 per hour for realtime, on self-serve and Business plans. Enterprise pricing is unchanged.
- Hosted MCP: 39 new tools build and publish ElevenAgents from any MCP client. Edits are staged in a draft, and only the publish tool makes them live.
- Tools as code: webhook, client and system tools can be built in the Python and JavaScript SDKs, from v2.79.0.

- Fast tier for Gemini: Gemini agent models accept `service_tier: fast` at priority rates, and a request is billed at the fast rate only when it is served at priority.
What should a team running voice agents check this week?
Make a short list and run it before December 15, 2026:
- Model per agent: list every agent and the model it uses. Anything on a deprecated model needs a new choice.
- Reasoning effort: set it on purpose for `gpt-6-luna` agents, so a default does not decide it.
- Test calls: place 10 to 20 calls on the new model and read the transcripts. Look at refusals, tool calls and call endings.
- Cost: check a day of calls against your bill. A new model can change the cost per minute.
- Owner: write down who watches the vendor changelog. Four models were deprecated in one entry.
Our guide to the ElevenLabs alternatives and the Retell, Topcalls and ElevenLabs comparison show how the platforms split these jobs.
Does the model underneath a voice agent matter to a sales team?
Yes, because the model decides what the agent says to your leads. A model swap is a script change you did not write. Any platform that lets you pick the model also hands you the upkeep of that choice. Ask each vendor three things: who picks the model, how much notice you get, and what happens to your agents on the shutdown date.
For a sales team, the cost side is simpler to compare. Topcalls AI calls start from $0.25 a minute, depending on the country called, billed by the second. The AI price includes the AI voice model, call recording, transcription and analytics, and there is no separate transcription bill. Each call returns a recording, a transcript, a summary, a disposition and a post-call analysis. See AI voice agents and smart campaigns.
Consent stays with you. Topcalls does not keep a do-not-call list: you own consent and opt-outs, and Topcalls controls dispatch hours and retries per campaign. To see what your own volume costs, run the ROI calculator or talk to us. Teams that qualify leads by phone can start from AI lead qualification.
Sources
- ElevenLabs changelog, November 17, 2026, ElevenLabs Documentation.
- ElevenLabs Speech to Text overview, ElevenLabs Documentation.
Frequently Asked Questions
Get AI calling tips in your inbox
No spam. One email per week with actionable sales automation tips.






