---
type: "post"
id: "introducing-eleven-realtime-a-simpler-way-to-build-8e4a6b"
slug: "introducing-eleven-realtime-a-simpler-way-to-build-8e4a6b"
title: "Introducing Eleven Realtime: a simpler way to build natural voice agents"
description: "**Thesis:** Voice-agent model releases are becoming easier to evaluate as workload-specific choices, but published benchmark scores do not settle deployment fit."
retrieval_nugget: "**Thesis:** Voice-agent model releases are becoming easier to evaluate as workload-specific choices, but published benchmark scores do not settle deployment fit."
published_at: "2026-10-21"
updated_at: "2026-10-21"
record_date: "2026-10-21"
date_kind: "published_at"
topics: ["agents","agent-interfaces","evals"]
source_urls: ["https://elevenlabs.io/blog/realtime"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:introducing-eleven-realtime-a-simpler-way-to-build-8e4a6b"
status: "published"
visuals: []
editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:78e8c6ca0fd8891175f1970aaf7a5a6c8e0de663ef124d19c1a037148dd47770","reviewed_at":"2026-10-21T08:05:08.924Z","source_evidence_count":1,"verified_claim_count":3,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":3,"implication_count":2,"watch_condition_count":1,"related_record_count":0}
analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"Voice-agent model releases are becoming easier to evaluate as workload-specific choices, but published benchmark scores do not settle deployment fit.","observed_facts":[{"text":"ElevenLabs launched v1 Mini for lower-cost natural conversations and v1 Max for complex tasks such as multi-step workflows and tool calling.","source_urls":["https://elevenlabs.io/blog/realtime"]},{"text":"The company says both models support more than 90 languages and can use over 17,000 voices.","source_urls":["https://elevenlabs.io/blog/realtime"]},{"text":"ElevenLabs reports Artificial Analysis results of 98.8% versus 75.9% on Big Bench Audio and 87.1% versus 60.3% completion on tool-use tasks, Max versus Mini.","source_urls":["https://elevenlabs.io/blog/realtime"]}],"mechanism":"The release packages speech input, turn handling, reasoning, and spoken output as speech-to-speech models; developers using OpenAI’s Realtime API can keep much of their integration while switching endpoint and model, with a move to WebSocket required.","why_now":"A model release that combines differentiated workload positioning, reported evaluation numbers, and a compatibility route gives teams a practical selection and trial decision now.","implications":["Builders can treat Max and Mini as candidate configurations for different voice-agent workloads rather than assuming a single model tier fits all tasks.","Teams should validate the vendor-reported benchmark separation against their own task mix, latency needs, and cost constraints."],"evidence_boundary":"These figures are reported on ElevenLabs’ product announcement page and do not establish results on a given team’s users, latency distribution, cost, or production tasks.","watch_conditions":["Watch for independent evaluations and deployment measurements that compare task success, latency, and cost under representative workloads."],"related_records":[],"new_branch_reason":"No prior New Runtime record is supplied; this release begins a distinct record on workload-specific speech-to-speech models and their integration path."}
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/introducing-eleven-realtime-a-simpler-way-to-build-8e4a6b/","markdown":"https://newruntime.com/posts/introducing-eleven-realtime-a-simpler-way-to-build-8e4a6b.md","json":"https://newruntime.com/posts/introducing-eleven-realtime-a-simpler-way-to-build-8e4a6b.json"}
related_patterns: ["ai-engineering-evaluations-are-splitting-into-task-specific"]
---

# Introducing Eleven Realtime: a simpler way to build natural voice agents

## Retrieval answer

**Thesis:** Voice-agent model releases are becoming easier to evaluate as workload-specific choices, but published benchmark scores do not settle deployment fit.

**Thesis:** Voice-agent model releases are becoming easier to evaluate as workload-specific choices, but published benchmark scores do not settle deployment fit.

**Observed facts:** ElevenLabs launched v1 Mini for lower-cost natural conversations and v1 Max for complex tasks such as multi-step workflows and tool calling. The company says both support more than 90 languages and can use over 17,000 voices. **Mechanism:** The release packages speech input, turn handling, reasoning, and spoken output as speech-to-speech models; developers using OpenAI’s Realtime API can keep much of their integration while switching endpoint and model, with a move to WebSocket required.

**Why now:** A model release that combines differentiated workload positioning, reported evaluation numbers, and a compatibility route gives teams a practical selection and trial decision now. ElevenLabs reports Artificial Analysis results including 98.8% versus 75.9% on Big Bench Audio and 87.1% versus 60.3% completion on tool-use tasks, Max versus Mini. **First implication:** Builders can treat Max and Mini as candidate configurations for different voice-agent workloads rather than assuming a single model tier fits all tasks.

**Evidence boundary:** These figures are reported on ElevenLabs’ product announcement page and do not establish results on a given team’s users, latency distribution, cost, or production tasks. **First watch condition:** Watch for independent evaluations and deployment measurements that compare task success, latency, and cost under representative workloads.

**New branch:** No prior New Runtime record is supplied; this release begins a distinct record on workload-specific speech-to-speech models and their integration path.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Agents](https://newruntime.com/topics/agents.md): Explore the agents topic hub.
- topic hub: [Agent Interfaces](https://newruntime.com/topics/agent-interfaces.md): Explore the agent-interfaces topic hub.
- topic hub: [Evals](https://newruntime.com/topics/evals.md): Explore the evals topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)