Best Free AI APIs for Vietnamese in 2026: Benchmark Scores and Free Tiers

The free API that scores best in Vietnamese today is Gemini 3.8 Flash on the Gemini API free tier: 78.40 on AI Singapore's SEA-HELM leaderboard, level with paid models such as Claude Opus 4.8 (78.70). Among open models, Qwen 3.5, Gemma 4 31B, GLM 5.2 and Qwen 3.6 are all within about two points of each other; Gemma 4 31B is the easiest to use for free, on the Gemini API, OpenRouter's free models and Requesty. The models built for Vietnamese, such as Vistral and fine-tunes of Sailor2, now score well below these general models, and we found no free hosted API for them.

Every score below comes from a published leaderboard, linked with its date; we read them on 2026-10-26 and did not run any test ourselves. Rows that link an offer page are free tiers in our verified catalog. Why models lose points in Vietnamese at all is explained in why AI is worse in Vietnamese.

Vietnamese scores and where to use each model free

SEA-HELM scores each model on Vietnamese tasks (knowledge, instruction following, multi-turn chat, generation, reasoning, understanding, culture and safety) and averages them into one Vietnamese score out of 100. The table lists the models with a free API in our catalog:

Source: SEA-HELM Vietnamese leaderboard (results dated 2026-09-18; 87 models: 71 open-weight of all sizes and 16 closed), our offer pages for the free limits, checked 2026-10-26. The leaderboard's default view shows only open models up to 200B, so the link above sets Model size to All and turns on closed models. SEA-HELM shows a 95% confidence interval of about ±1.5 points for the top models, so models within two points of each other are effectively tied.

GLM 5.2 (76.93) is also near the top, but Cloudflare lists it among the models that need Workers Paid, and we found no other free API for it.

Gemini: the strongest free option

With closed models turned on, SEA-HELM's Vietnamese table puts Gemini 3.8 Flash (78.40) within its confidence interval of the best paid models: Claude Opus 4.8 (78.70), Gemini 3.1 Pro Preview (78.66, not on the free tier) and GPT 6 Astra (77.86). Gemini 3.6 Flash (77.62) and Gemini 3.1 Flash-Lite (75.24) are also on the free tier.

The Gemini API free tier is open in Vietnam, needs no card, and also serves Gemma 4. The catch is that Google shows the free per-model limits only inside AI Studio, not on a public page, so check your quota there before you build on it. Steps: how to get a free Gemini API key.

Models made for Vietnamese

VMLU is a Vietnamese multiple-choice benchmark of 58 subjects, most with about 200 questions, from elementary school to professional level, grouped into STEM, social science, humanities and others. Its leaderboard shows the Vietnamese-tuned open models behind the general ones evaluated at the same time:

Source: VMLU leaderboard, Sailor2-8B-Chat model card, Vistral-7B-Chat model card, checked 2026-10-26. VMLU lists no row for Sailor2-8B-Chat itself, only this fine-tune of it. The top of VMLU's fine-tuned table (entries up to 2026-10-23) is private models from Vietnamese companies such as VAI, VinSmart Future and MISA, with no free API, so they are left out.

Sailor2 and Vistral are still useful if you need a small model you can run yourself: download the weights from Hugging Face and run them locally (how to run AI locally for free), or fine-tune one on your own Vietnamese data (fine-tune an LLM for $0). For an app on a free API, a general model from the first table will usually answer better.

SEA-LION: Southeast Asian models with a free key

AI Singapore's SEA-LION models are trained further on Southeast Asian languages. The SEA-LION API gives one free trial key per Google account, at 10 requests a minute, with models including Qwen-SEA-LION v4.5 27B (73.85 on SEA-HELM Vietnamese, function calling supported) and Llama-SEA-LION v3.5 70B for reasoning. AI Singapore publishes no price or end date for the trial key. It is not yet in our catalog.

Steps: sign in to the SEA-LION Playground with Google, open the API Key Manager, click Create New Trial API Key and save the key (it is shown once), then call https://api.sea-lion.ai/v1 with any OpenAI-compatible client.

Source: SEA-LION API guide, checked 2026-10-26.

How to choose

• Best free quality in Vietnamese: Gemini 3.8 Flash on the Gemini API free tier.
• Best open model: Gemma 4 31B on the Gemini API free tier or OpenRouter. Keep a second provider configured, since OpenRouter's free models allow only 50 requests a day until you buy credits.
• Many requests a day: GPT OSS 120B on Groq (1,000 requests a day), about seven points lower in Vietnamese than Gemma 4 31B.
• Southeast Asian focus or function calling in Vietnamese: the SEA-LION API.
• Your own small model: Sailor2 or Vistral, run locally.

Whatever you pick, test it on 20 or 30 of your own Vietnamese prompts first. A leaderboard average hides weak spots: on SEA-HELM, Gemma 4 31B and Qwen 3.5 122B differ by less than a point overall but by about five points on Vietnamese safety.

Limits and the catch

• Vietnamese costs more tokens. The same sentence takes more tokens in Vietnamese than in English, so per-token free limits go less far: why other languages use more tokens.
• Benchmarks are not your task. SEA-HELM and VMLU measure exams and generic tasks; a model that scores two points higher may still be worse at your domain.
• Free tiers may train on your data. Google may use Gemini API free-tier prompts to improve its products; keep personal data out of free tiers unless the provider says it does not train on them.
• Scores and free models change. Leaderboards add models and free lineups rotate. We re-check monthly; the date at the top of this page is our latest check.