brain/
conceptartificial-intelligence

Gemini 3.5 Transcribe (August 2026)

Notes

Gemini 3.5 Transcribe (August 2026)

Vintage: 2026-08. Primary evidence is one official @GoogleDeepMind X post dated 2026-08-26 (text/photos). Video announce (…9221477077101) was held for /transcribe-clipping and is not in the source Items. Linked goo.gle URL was not fetched. Capability list is as written — do not invent SKU or pricing. 2026-09-15 attach: Live companion restates last-month release + issuer WER — not a new Transcribe launch.

One-line summary: Google DeepMind listed Gemini 3.5 Transcribe capabilities — better handling of complex phone numbers, postal codes, and order IDs in noisy environments; filler-word removal and auto-format; custom vocabulary for unique names and product titles; automatic speech detection and transcription in 85+ languages — and said to try it in the Gemini app on macOS and Gboard on Android, and to start building in Google AI Studio and Antigravity.

The insight

This is a lab-official X capability list, not a gemini-3-7-flash model card and not a fetched product page. Last30days (2026-08-27-significant-ai-developments-last-30-days) attached a pointer-only note on gemini-3-7-flash (post text not fetched; no Transcribe page). This page is the primary-X ingest. Distinct from gemini-macos-voice (@GeminiApp dictate/summarize/rewrite — a different product-account post, no Transcribe name).

Evidence

The bullets are one X post (fetch_method: x-mcp) in 2026-08-27-x-ai-news-27-aug-2026-deepmind-double-blind-evals. Permalink lives on the source page.

  • From 2026-08-27-x-ai-news-27-aug-2026-deepmind-double-blind-evals (official @GoogleDeepMind, 2026-08-26): "It’s better at understanding complex phone numbers, postal codes, and order IDs – even in noisy environments."
  • From the same post: "It can remove filler words and auto-format your text."
  • From the same post: "It recognizes custom vocabulary so the model knows unique names and product titles."
  • From the same post: "It automatically detects and transcribes speech in 85+ languages."
  • From the same post: "Try it now in the @GeminiApp on macOS and Gboard on @Android."
  • From the same post: "Start building in @GoogleAIStudio and @Antigravity." Linked t.co/lNlILGYFVy was not fetched. Photo attachments. Video announce held for /transcribe-clipping — not cited.
  • From 2026-09-08-x-afternoon-openai-navier-stokes-images-2-5 (@GeminiApp / follow-up, 2026-09-08): Gemini 3.5 Transcribe in the Gemini macOS app — product push. Model announce was 2026-08-26 (Google blog). Do not treat as today’s model drop.
  • From 2026-09-15-google-gemini-3-8-live-3-8-live-extended-thinking (Build real-time voice applications with Gemini 3.8 Live and 3.5 Transcribe, Sep 15, 2026): dedicated STT released last month; average WER 4.0% streaming and 2.6% non-streaming; 85+ languages; custom vocabulary biasing (up to 1,000 terms); smart transcription mode; Interactions API for files up to 1 hour with timestamps/speaker labels. Do not date Transcribe as a Sep 15 launch. WER / 1,000-term / 1-hour figures are issuer restatement on the Live companion, not a first-party Transcribe model-card parse.

What this source does not establish

  • No fetched original Transcribe model card. The August goo.gle URL was not fetched. The Sep 15 companion restates WER / vocab / Interactions API; that is still not a parse of the August Transcribe card.
  • No SKU or plan gate in the August X post. Do not invent Pro / Ultra / free availability. The Sep 15 companion names Interactions API for files up to 1 hour — cite that post, do not back-fill it onto the August X thread.
  • Not a gemini-3-7-flash claim. Do not fold Transcribe into the Flash-tier page as a Flash feature.
  • Video announce was not transcribed and is not a source.
  • "85+" is as written. Do not invent a language list.

Open questions

  • What does the unfetched August linked page add (model card, pricing, API surface) that the X post omits? Sep 15 companion adds issuer WER 4.0% / 2.6% and Interactions API — still not that August card.
  • What does the held video announce say that the capabilities reply does not? Left for /transcribe-clipping.

Related

Referenced by