[
    {
        "role": "user",
        "content": "Can%20you%20tell%20me%20what%20tts%20engine%20the%20buttery%20voice%20is%20using%20on%20the%20clude%20app%20for%20ios%3F"
    },
    {
        "role": "assistant",
        "content": "I couldn\u2019t find any credible public information that confirms what specific TTS engine the \u201cButtery\u201d voice uses in the Claude app (iOS). The official support documentation doesn\u2019t disclose details about the internal TTS vendor or model names. ([support.claude.com](https://support.claude.com/en/articles/11101966-use-voice-mode?s=09&utm_source=openai))\n\nIf you like, I can try to dig deeper (e.g. developer statements, code inspect, etc.) to see if there\u2019s more low-level info. Would you prefer I try that?"
    },
    {
        "role": "user",
        "content": "yes%20please"
    },
    {
        "role": "assistant",
        "content": "I dug around but didn\u2019t find a definitive source confirming exactly which text-to-speech (TTS) engine powers the *Buttery* voice in Claude for iOS. However, there are some credible leads and clues worth sharing, so you can judge how likely they might be.\n\n---\n\n### \ud83d\udd0d What we *do* know\n\n1. **Voice mode and model update**  \n   In July 2026, Anthropic updated Claude\u2019s voice mode. Originally it was powered by the *Haiku* model; now it \u201cpicks the last model people used in text chat and uses its fastest version by default,\u201d meaning voice outputs can come from models like *Haiku*, *Opus*, or *Sonnet*. ([techcrunch.com](https://techcrunch.com/2026/07/23/anthropic-updates-claude-voice-mode-with-more-capable-models/?utm_source=openai))\n\n2. **Anthropic\u2019s Transparency Report**  \n   The reports say Claude can output *audio via text-to-speech* as one of its modalities. But the documentation does *not* name a third-party TTS provider for these voices, or give internal model names tied to things like \u201cButtery.\u201d ([anthropic.com](https://www.anthropic.com/transparency/model-report?utm_source=openai))\n\n3. **User observations and crowdsourced info**  \n   On Reddit, some users report that Claude\u2019s voice mode \u201cuses ElevenLabs technology,\u201d and that Anthropic lists ElevenLabs as a subcontractor. Others are comparing the \u201cButtery\u201d voice to voices from ElevenLabs, or saying the accent/style is close. These are *user claims*, however\u2014not confirmed by a primary source. ([reddit.com](https://www.reddit.com/r/ClaudeAI/comments/1ugaf67/claude_readloud_voice_change/?utm_source=openai))\n\n---\n\n### \u26a0\ufe0f Caveats / what we *don\u2019t* know\n\n- There\u2019s **no official public confirmation** from Anthropic (in their Help Center, blog, or developer docs) that \u201cButtery\u201d is powered by ElevenLabs or any specific external TTS vendor.\n- Reddit threads are anecdotal and subject to misinterpretation or hearsay.\n- Voice style (tone, cadence, accent) similarities don\u2019t always map to a shared engine; different TTS systems can mimic certain stylistic features or use voice models that sound alike.\n\n---\n\n### \ud83d\udc4d Likelihood assessment\n\nHere are how possible scenarios stack up, given the evidence:\n\n| Hypothesis | Support | Weakness |\n|------------|---------|----------|\n| **\u201cButtery\u201d uses an external TTS vendor like ElevenLabs** | Users report as much; ElevenLabs is named as a subcontractor in some contexts. ([reddit.com](https://www.reddit.com/r/ClaudeAI/comments/1ugaf67/claude_readloud_voice_change/?utm_source=openai)) | No official confirmation; could be mistaken association. |\n| **\u201cButtery\u201d is generated via Anthropic\u2019s in-house models (Haiku/Sonnet/Opus)** | Voice mode now leverages those for voice responses. ([techcrunch.com](https://techcrunch.com/2026/07/23/anthropic-updates-claude-voice-mode-with-more-capable-models/?utm_source=openai)) | Doesn\u2019t explain user mentions of external vendor; voice style might be trained internally. |\n| **Hybrid or mixed approach (in-house model + third-party tools/components)** | Often companies combine internal and external tech; subcontractor mention supports this possibility. | No documents describing such a hybrid architecture for \u201cButtery.\u201d |\n\n---\n\n### \u2705 My best guess\n\nBased on the current evidence:\n\n- It\u2019s *plausible* that \u201cButtery\u201d is powered by Anthropic\u2019s internal voice/TTS stack (Haiku, Sonnet, or Opus models), especially since those are now used for voice mode.  \n- But there\u2019s a **non-trivial chance** that Anthropic also leverages **ElevenLabs** for at least some voices or as a component (e.g. voice model licensing or voice style transfer), given the Reddit mentions and subcontractor reference.  \n\n---\n\nIf you like, I could try to see if there\u2019s leaked technical data (e.g. via patents, job postings, or source-code permissions) to get a more concrete answer. Do you want me to chase that kind of deep dive?"
    }
]