Gemini Flash-Lite TTS lacks 29 languages that Flash covers

On Google's text-to-speech page, Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS share 103 languages, and 29 more are marked as Flash only. Thai, Swedish and Finnish are among the Flash-only ones, so the cheaper Flash-Lite model cannot be assumed to cover a language just because Flash does. MAI-Voice-2.1 lists 23 languages, shown as 28 locale entries, and it names Thai, Swedish and Finnish on the page.

The count and five checks

I counted the rows of the table in Google's speech-generation page: 132 languages, 103 with a check mark in both columns and 29 with a check for Flash and a dash for Flash-Lite. The page's own wording is "over 130" for Flash and "over 100" for Flash-Lite, which agrees. The Microsoft page lists 23 languages and 28 entries because some languages appear with several locales (four English, two Spanish, two Portuguese).

The 29 Flash-only languages

The languages that the page marks for Flash only are: Banjar (Arab script); Bashkir; Bemba; Burmese; Crimean Tatar; Dyula; Dzongkha; Finnish; Guarani; Igbo; Kabyle; Latgalian; Lithuanian; Luxembourgish; Minangkabau (Latn script); Occitan; Pangasinan; Sindhi; Slovenian; Somali; Southern Sotho; Swahili (individual language); Swati; Swedish; Tajik; Thai; Tigrinya; Tosk Albanian; Uyghur.

• Several of these are widely used languages in commerce (Thai, Swedish, Finnish, Lithuanian, Slovenian), so check the table row before you pick the lower price.
• The Gemini page says the model detects the input language automatically, so there is no language switch to flip if a language is missing; the result for an unsupported language is something to test, not assume.

What I can and cannot say about Sume

Sume's TTS job records a language field, as the jobs page documents, and its STT accepts a language_code hint or auto-detects (Video inspect). I found no language count in the Sume docs or the catalog entries, so I give none. Before you rely on a language, run one short line through the Sume job and read back the result, rather than inferring from a vendor's table.

Price against coverage

The reason to check is price. On Google's pricing page, Flash-Lite TTS output is $6.00 per million audio tokens to the end of 2026 and Flash is $9.00, so Flash-Lite is one third cheaper per audio minute ($0.009 against $0.0135, from 1,500 tokens a minute). A team that picks Flash-Lite for a 12-language campaign saves 33.3 percent on audio output, but only if all 12 languages sit in the shared 103. If one of them is Thai or Finnish, the saving is gone, because that language needs Flash.

From January 1, 2027 the listed prices double, to $12.00 and $18.00, so the ratio stays at one third while both bills grow.

• Step 1: list the languages you ship, by the names used in Google's table.
• Step 2: mark any that appear in the Flash-only list above.
• Step 3: if none do, price Flash-Lite; if any do, price Flash for those languages only and Flash-Lite for the rest, which means two model ids in your integration.