What changed
ElevenLabs’ changelog documents several agent and API changes, including newer default templates using scribe_realtime instead of the removed elevenlabs ASR provider, expanded agent-testing limits, and new controls for turn detection and soft-timeout filler messages.
The same documentation lists deprecations for the eleven_monolingual_v1 and eleven_multilingual_v1 text-to-speech models and for scribe_v1 speech-to-text. The changelog directs users toward eleven_multilingual_v2, current speech models or scribe_v2/scribe_v2_realtime as replacements.
Why it matters
This is a maintenance and migration update more than a consumer-facing feature launch. Teams with older voice or transcription integrations should inventory model IDs, SDK versions, agent templates and fallback behavior before a removal date affects production.
Checks before publishing
- Confirm the exact removal date in the current changelog and migration documentation.
- Run a compatibility test with the replacement model; voice, latency, punctuation and cost may differ.
- Review agent defaults after SDK upgrades rather than assuming old provider settings remain active.
- Do not treat higher test limits as a guarantee of higher production quotas.
Primary source: ElevenLabs documentation changelog.
Our editorial read
A leading voice platform for expressive speech, dubbing and agent audio workflows.
Best fit
Creators and product teams that need natural-sounding voices across languages and can manage consent and rights.
Where we would hesitate
Do not clone or publish a person’s voice without documented permission, and check model migration notices before shipping.
Decision rule
Choose ElevenLabs for voice quality and breadth; choose a broader editor when audio is only one small part of the workflow.
Evidence note: This judgment synthesises the linked official documentation, release notes and product context. It is not a claim that AI Tools Manual independently benchmarked every tool.
August 2026: hosted MCP, CLI and Procedures
ElevenLabs’ August documentation describes a hosted MCP server at https://api.elevenlabs.io/v1/mcp, authenticated through OAuth, as the replacement for the deprecated local MCP server and MCP player. The August 24 release also introduces an ElevenLabs CLI that exposes API operations as commands and can manage agent configurations, branches, tests, UI components, and data-residency selection. Procedures are now generally available in ElevenAgents for loading task-specific instructions when a trigger matches.
For teams, this changes both the integration path and the control surface. Confirm which client can connect to the hosted server, what OAuth scopes it receives, and how credentials, logs, tests, and regional residency are governed. Treat Procedures as executable policy: keep triggers narrow, version changes, and review a test run before enabling external actions. The release notes are the source for these capabilities; we have not independently tested the CLI, MCP server, or agent limits.
Primary sources: ElevenLabs August 24 changelog and August 22 hosted MCP notice.
September 2026 update: ElevenLabs adds Music v2.5 and a more scriptable platform
ElevenLabs says Music v2.5 is now the default model for prompted and reference generation in ElevenMusic. The company describes more layered arrangements and natural-sounding instruments, while keeping the earlier Music v2 available. Music v2.5 is also listed in ElevenCreative and the ElevenLabs API, so a team can score a video in Studio, run Music as a node in Flows, or call the model from its own product.
The announcement includes concrete usage terms that matter more than the “best model” headline: ElevenLabs says creators own tracks made in ElevenMusic on every plan, including Free. Free is listed with five lossless downloads per day, while Pro is listed with 400 lossless downloads per month. Commercial use on Free requires crediting ElevenMusic, and downloads are blocked when a track references another artist’s song. Re-check the current plan page and terms before publishing client work; these limits and licensing conditions can change.
What the September changelog means for teams
The September 7 changelog adds the elevenlabs say CLI command, makes eleven_v3 the default TTS model in the CLI, and adds workspace-wide conversation ticket retrieval for accessible agents. Together with the August hosted MCP server, CLI configuration and generally available Procedures already described in this ElevenLabs review, the platform is moving toward versioned, repeatable voice and agent operations rather than one-off web experiments.
Suitable users: video and podcast teams that need a repeatable music-to-voice pipeline, developers who want API/CLI automation, and support teams that need to triage conversations across multiple agents. Less suitable: projects that require guaranteed artist-style imitation, unlimited downloads, or a fully self-hosted audio stack. The music announcement explicitly blocks tracks referencing other artists’ songs, and we have not independently tested the model’s musical quality, latency, API reliability or rights enforcement.
Our decision rule is to prototype one short, rights-cleared asset, record the model version, plan, prompt, credits and export count, then have a human review attribution, originality, lyrics, brand safety and downstream licensing. Keep the old Music v2 path available until a real project passes your acceptance checklist. For a broader workflow choice, compare the voice and editing alternatives in the site’s comparison guides.
Evidence note: model behavior, ownership language, download allowances and CLI changes are ElevenLabs’ own claims from its product announcement and official changelog. They are not independent performance measurements or legal advice.
Primary sources
- ElevenLabs: Introducing Music v2.5
- ElevenLabs Documentation: September 7, 2026 changelog
- ElevenLabs pricing and plan terms
September 23, 2026 migration alert: Sora 2 is retired in ElevenLabs Image & Video
ElevenLabs’ official changelog says Sora 2 and Sora 2 Pro were removed from the Image & Video model picker and stopped generating on September 24, 2026. Existing generations remain in history, but Flows and templates that contain a Sora node must be switched to another video model, such as Gemini Omni 1.1 Flash, before they can run again. The same entry says Seedance 1.5 Pro is deprecated and will stop working on November 11, 2026, with Seedance 2.0 named as the replacement. These are provider migration notices, not an independent quality comparison.
September 21, 2026 agent and image API changes
The changelog also adds parallel tool calls for ElevenAgents, with the enable_parallel_tool_calls setting enabled by default, and adds gpt-6-astra to the agent model options. Agent alerting now supports Slack as well as PagerDuty and webhooks. On the image side, gpt-image-2.5-flare and gpt-image-2.5-sunburst accept up to ten reference images, 1K/2K/4K output, fourteen fixed aspect ratios plus auto, and quality levels through max. The documented API details are useful for planning, but they do not prove that parallel calls improve a particular production workflow.
Who is affected and what to do
Teams using ElevenLabs Flows should inventory Sora and Seedance nodes, copy existing outputs, and test the replacement model with the same prompt, duration, aspect ratio, and approval checklist. Voice-agent teams should decide whether parallel tool calls are safe for their integrations, add idempotency to side-effecting actions, and route Slack alerts to a monitored channel. Projects with clinical audio or sensitive customer data should separately verify the Scribe v2 Medical terms, retention, and data residency before adoption.
Our assessment and limits
The immediate decision is migration readiness, not whether ElevenLabs has the “best” image model. ElevenLabs remains a strong fit for teams already using its voice, agent, and media stack; it is not a fit for a workflow that requires a frozen video-model catalog or assumes a deprecated node will continue to run. We have not independently tested the replacement video models, parallel execution, GPT-6 Astra, Slack alerting, or the image variants. Re-check the official ElevenLabs changelog and product terms before publishing a client asset.
For context, read our ElevenLabs review, compare ElevenLabs with Descript, and use the dubbing review guide. The official changelog is the primary source for the retirement dates and API fields; our recommendations are editorial guidance rather than a benchmark.
September 23, 2026: model retirements affect existing video workflows
ElevenLabs’ changelog states that the Sora 2 and Sora 2 Pro APIs were discontinued on September 24, 2026. The models were removed from the Image & Video picker and stopped generating on that date, while previously created generations remain in history. The changelog specifically warns that Flows and templates containing a Sora node must be switched to another video model before they will run again. It also says Seedance 1.5 Pro is deprecated: it is no longer offered for new generations and is scheduled to stop working on November 11, 2026; ElevenLabs points users to Seedance 2.0. These dates are operationally important for teams with saved automations, not just new buyers choosing a model.
Migration checklist for affected creators and teams
Search existing Flows, templates, and API requests for retired model identifiers. Duplicate each affected workflow before editing, replace the model node, then validate aspect ratio, duration, reference inputs, audio behavior, and output consistency. Re-run representative prompts and check whether the new model changes credit usage or licensing assumptions. Keep a fallback workflow and export any required assets from generation history while they remain accessible. Do not assume a replacement model is a drop-in equivalent: the changelog identifies options, but does not promise identical quality or behavior.
Evidence and editorial judgment
This is a compatibility and maintenance issue; creators depending on existing templates should prioritize migration before exploring newer models. Primary source: the ElevenLabs official changelog. Compare alternative generation workflows in our Runway review and Descript review. AI Tools Manual did not test migration or verify a specific account’s saved workflows. Confirm current model availability, output history, billing, and any account-specific deprecation notice before treating a workflow as restored.
October 5, 2026: ElevenLabs exposes Flows for image and video automation
ElevenLabs says image generation, video generation, and templates built in its Flows canvas are now available through the ElevenLabs API. The company describes templates as reusable workflows that can combine image, video, voice, music, and sound-effects models. A team can provide new inputs—such as a product image and name—to a saved template rather than rebuilding the composition and prompts for each item. The announcement also says the image and video models and Flows nodes can be used through the ElevenLabs MCP from Claude, ChatGPT, or Cursor. It names models from Google, OpenAI, ByteDance, and Creatify; this is a vendor-stated integration list, not an independent quality ranking.
The practical shift is from creating individual assets in a user interface toward triggering repeatable creative production from another system, such as a product catalog or storefront. That may help teams handling many SKUs or channel variants, but automation can also multiply a flawed prompt, inaccurate product depiction, or unapproved claim. The announcement does not establish that every model, connector, template, or integration is available to every account or region.
How to evaluate the API workflow before scaling
Start with a small, non-public batch. Check whether the API and MCP expose the exact Flow nodes your workflow needs; compare generated assets with the supplied product details; inspect crop, text, brand references, voice, and output format; and have a human approve each template before it is connected to a live catalog. Keep a record of the template version, inputs, outputs, and reviewer so that a changed model or prompt can be traced. Confirm current API access, plan eligibility, usage accounting, data handling, and output rights in the relevant documentation and account terms before sending customer or unreleased-product data.
Our assessment: the announcement makes ElevenLabs more relevant to teams automating multimodal creative operations, not just voice production. It is less compelling for a solo creator who only needs one-off image or video generation and does not need an API pipeline. We have not connected an account, benchmarked output quality, or verified production reliability. Teams comparing speech and dubbing workflows can also review ElevenLabs vs. Descript and our AI dubbing review guide.
Primary source: ElevenLabs, “Introducing the Image, Video, and Template APIs,” published September 26 and updated October 5, 2026. Verify current API documentation, account access, data terms, and usage details before implementation.
October 5, 2026: ElevenAgents procedures and merge proposals
ElevenLabs’ October 5 changelog describes a workflow change for structured procedures: saving or publishing an agent now compiles and validates those procedures automatically. The Update Agent and Create Agent draft flows return per-procedure validation errors when a procedure is invalid. The changelog labels the separate Compile Procedures endpoint as legacy, while saying it remains available to existing callers. This is an API lifecycle signal, not a published removal date.
The same changelog lists merge proposals in alpha. Teams can open an agent-branch proposal, review configuration changes and test results, leave comments or approval decisions, and merge an approved proposal into a target branch. JavaScript and Python SDK v2.71.0 also add clients for agent merge proposals, deployment history, test-invocation cancellation, and Speech Engine duplication. These additions matter most to teams operating agents as versioned, tested systems rather than to creators who only generate one-off speech.
Migration and fit: treat this as a developer workflow update
Best fit: developers and teams managing structured ElevenAgents procedures across branches, with a review process for changes. It is less relevant to a voice creator who does not use the agent API. Our assessment is to test one non-production agent first: confirm procedure validation in the save/publish flow, inspect the exact error payloads, and compare the result with any existing compile-endpoint integration. Because the endpoint is only marked legacy and no retirement date is stated in the cited note, do not remove working calls until the current API documentation and support guidance are checked.
Merge proposals are explicitly described as alpha, so teams should not treat them as a mature replacement for an established release-control process. Keep a human approval step, retain a rollback path, and test deployment history and cancellation behavior before relying on the SDK clients in production. For broader product context, compare this agent workflow with our ChatGPT review. Evidence note: this summary is based on the official ElevenLabs changelog; no production integration was tested. Next review: October 15, 2026.
October 12, 2026: ElevenLabs expands agent queueing, voice collections and audio effects
The official ElevenLabs changelog for October 12 documents several developer-facing changes. Newly created ElevenAgents now have call queueing enabled unless the configuration explicitly disables it; existing agents keep their current setting. The release also adds ticket merging and search, success/failure outcomes for workflow end nodes, retry support for post-call audio webhooks (with documented size caps), voice-collection management through the API, and optional audio effects for Text to Speech and Text to Dialogue. ElevenLabs states that audio effects do not change request cost. See the official changelog.
Decision impact: teams creating new voice agents should check queueing defaults and configure wait behavior intentionally rather than assuming a new agent matches an older one. Builders of support workflows may also benefit from ticket search/merge and explicit end outcomes, while creative API users can explore the audio-effects catalog. This is an API and workflow update; it does not show that speech quality improved.
We have not run migration tests or independently verified billing. The docs describe behavior, limits and defaults, but individual agents, models and output formats may have different support. This is most useful for developers operating voice workflows; a casual creator may see little change. For adjacent audio creation options, see Suno and CapCut. Next review: November 16, 2026.
October 19, 2026: Eleven Realtime and Realtime Translation
ElevenLabs added Eleven Realtime, a family of complete voice-agent configurations. Its changelog names a default Mini variant and a Max variant for complex instructions and multi-step tool calls; the provider says an agent’s prompt, voice, tools, and knowledge base continue to work with the new configuration. This is a product/API change, so developers should confirm the model identifier, workspace rollout, supported transport, and any migration steps in the current docs before changing a deployed agent.
Realtime Translation is enterprise-scoped
The same release describes Realtime Translation as speech-to-speech translation while the speaker is still talking, exposed through a WebSocket API. ElevenLabs says it is available for Enterprise workspaces. Its documentation also describes Zero Retention Mode for keeping transcripts and translations out of logs and history. Both the access statement and the precise retention behavior need review against the account’s contract and live API configuration; do not infer that “zero retention” applies automatically to every workspace or data path.
Editorial assessment: this could matter to teams building low-latency voice agents or live multilingual experiences, but it is not yet evidence of general consumer access or a measured quality advantage. Start with a non-sensitive pilot, check end-to-end latency and language coverage on representative audio, and preserve a rollback path. AI Tools Manual did not test the models or translation quality.
Primary source: ElevenLabs changelog and the Realtime Translation documentation. Next review: November 23, 2026. Human review: confirm Enterprise eligibility, model/API identifiers, migration requirements, supported languages, and retention configuration.