Introduction

Speechify's Simba 3.5 text-to-speech model is now on OpenRouter. It went live there on November 11, 2026, as speechify/simba-3.5. It's a closed, API-only streaming TTS model: one model id speaks 48 languages, and you call it through OpenRouter's OpenAI-compatible /api/v1/audio/speech endpoint for $15 per 1M characters. If you already have an OpenRouter key for LLMs, that same key now gets you Simba 3.5 voice output. Simba 3.5 itself first shipped on Speechify's own API on October 21. What's new this week is wider availability, plus two deadlines on Speechify's direct API: the launch discount ends November 21, and the legacy Simba 1.6 ids switch over to 3.5 that same day.

Sources: OpenRouter's Simba 3.5 model page (listed as released November 11, 2026) and TTS guide, the

(November 11 UTC) and
(November 12), Speechify's Simba 3.5 announcement (October 21, updated November 9), and the Speechify API changelog (November 9 and 10 entries).

What Shipped

  • Model: Simba 3.5 by Speechify (the SpeechifyAI developer platform). Text-to-speech, streamed audio.
  • Open or closed: closed. No weights. You can only reach it through Speechify's API or OpenRouter.
  • New on OpenRouter: model id speechify/simba-3.5 (canonical speechify/simba-3.5-20261111). Speechify is the only provider, so there's no routing choice. OpenRouter's models API lists 20 voices for it, including geffen_32, alec, alicia, archie and beatrice_32.
  • Price on OpenRouter: $15 per 1M input characters. Output is free.
  • Languages: 48 under one id, from Arabic and Bengali to Tagalog, Urdu and Vietnamese. Regional variants count, so pt-BR and pt-PT are both "Portuguese". That's up from 6 on Simba 3.0.
  • What OpenRouter does not expose (yet): the endpoint record shows supports_voice_cloning: false, and supported_parameters is empty. There's no cloning, and there are no OpenRouter instructions style controls for this model. OpenRouter's TTS guide also doesn't document a Speechify language option yet, so start with voices native to your target language.

On Speechify's Own API

  • Endpoints: simba-3.5 works on POST /v1/audio/speech, POST /v1/audio/stream, POST /v1/audio/stream/with-timestamps (word timings) and the WebSocket at GET /v1/audio/stream/ws. Set language (for example de-DE) when your text isn't in the voice's own language. Without it, the voice's locale decides.
  • Voices: any catalog voice can read any of the 48 languages, and so can a voice you cloned on a paid plan. A voice keeps the accent of the recording it came from, which is why Speechify flags "staff pick" native voices per locale in GET /v1/voices.
  • SSML: the break, prosody rate and sub alias tags carry over from Simba 3.2. Since October 20, prosody pitch and prosody volume are applied too.
  • Pricing (November 9 cut): list prices dropped from $20 / $16 / $12 to $15 (Free and Starter), $12 (Pro) and $10 (Scale) per 1M characters. Until November 21, Simba 3.5 still bills at the Simba 3.0/3.2 launch rates of $10 / $8 / $6. From November 22 it bills at the new list price.
  • Simba 1.6 sunset (November 10 entry): from November 21, requests that still name simba-english or simba-multilingual are served and billed as Simba 3.5 in every language. Nothing errors, but the audio changes. To stay on cheaper English-only audio, pin simba-3.2 before that date.
  • Multi-speaker: dialogue scripts stay on simba-dialogue-1.0 (POST /v1/audio/dialogue), not Simba 3.5.

Quick Start on OpenRouter

The request shape follows OpenRouter's TTS guide. The response is raw audio bytes, not JSON:

curl https://openrouter.ai/api/v1/audio/speech \
  -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "speechify/simba-3.5",
    "input": "One model, forty-eight languages. Ready when you are.",
    "voice": "geffen_32",
    "response_format": "mp3"
  }' \
  --output simba.mp3

  • response_format defaults to pcm on OpenRouter. Ask for mp3 if you're saving a file to play back, and use pcm for low-latency pipelines.
  • Always pass a voice. OpenRouter rejects requests without one for providers that don't have a documented default voice.
  • OpenRouter's model page has a benchmark panel with tricky prompts (invoices, units, code-switching, bracketed [whispering] cues, an embedded "ignore the previous text" line). Listen to those before you ship. Simba 3.5 does not list cue or instruction support, so expect bracketed cues to be read as text or ignored.

What Speechify Claims

These numbers come only from Speechify's fact sheet, and nobody has checked them independently yet:

  • 63% win rate over Simba 3.2 in blind side-by-side English listening tests (95% CI 60–67%), with US and UK listeners.
  • 62% preference over Simba 3.0 Multilingual in the six locales both models cover.
  • 72 ms time to first audio for 95% of streaming requests under production load.
  • 1.8× Simba 3.2's speech throughput per serving instance (1.4× vs Simba 3.0 Multilingual).

Why It Matters

OpenRouter's speech catalog now has 39 TTS models in one place. Simba 3.5 enters at the same $15 per 1M characters as Fish Audio S2.1 Pro and AssemblyAI TTS-1. It's above Voxtral Mini TTS 3.0 ($13) and well under Deepgram Aura-2 ($30), Eleven v4 ($80) and MiniMax Speech 2.8 HD ($100), based on OpenRouter's list rates today. For builders already routing LLM calls through OpenRouter, one key and one bill can now cover a 48-language reader voice, with no new vendor account needed.

Caveats:

  • Going direct is cheaper for now. Until November 21, Speechify's own API bills Simba 3.5 at $10 per 1M on Starter, versus $15 on OpenRouter. After that, Starter list and OpenRouter match at $15, and Pro and Scale plans are cheaper going direct.
  • On OpenRouter you get a 20-voice subset with no cloning and no word timestamps. The full voice catalog, cloned voices, speech marks and WebSocket streaming are on Speechify's API.
  • If you have old integrations on simba-english or simba-multilingual, they'll sound different from November 21. Test now, or pin simba-3.2.

Conclusion

Simba 3.5 on OpenRouter is a distribution story, not a new model: Speechify's 48-language streaming TTS is now one model id away for anyone with an OpenRouter key, at $15 per 1M characters. If you call Speechify directly, the dates to remember are November 21 (launch pricing ends, Simba 1.6 ids reroute to 3.5) and November 22 (new list prices apply). Try the OpenRouter benchmark clips, compare a native voice in your target language, and pick the route that fits your bill and feature needs.

—Anabel ♡