Bilibili matters to AI buyers because a NASDAQ-listed video platform also ships open translation and speech models. Index-Translate-9B runs on your own hardware with a 24.9k-star sibling, IndexTTS, covering voice cloning, so check it first when you need self-hosted translation or speech.
Bilibili is a Shanghai-based video community founded in 2009 and listed on NASDAQ as BILI. Its Index team publishes open models on Hugging Face, including Index-Translate, a family covering 150 languages with an Apache-2.0 9B version, and the IndexTTS voice-cloning system.
Founded: 2009 · HQ: Shanghai, China · CEO: Rui Chen · Funding: Public (NASDAQ: BILI; HKEX: 9626)
About Bilibili
Bilibili is a Chinese video community that launched its first website in June 2009 and took the name "bilibili" in January 2010. Its community spans lifestyle, games, entertainment, anime and tech content. Rui Chen is Chairman and CEO. The company is headquartered in Shanghai and trades on NASDAQ (BILI) and the Hong Kong exchange (HKEX 9626). Most of its revenue comes from mobile games, value-added services, advertising and e-commerce, per its investor relations site, so it is a video platform first and an AI lab second.
The AI work sits with the Index team, which publishes openly as IndexTeam on Hugging Face (22 listed team members on 2 November 2026). Its newest line is Index-Translate, a translation family built on the Qwen3.5 base from Alibaba Cloud and released in three sizes: Index-Translate-2B, Index-Translate-9B and a 35B-A3B preview. The 9B model is released under Apache-2.0, supports text translation across 150 languages and also translates text inside images without a separate OCR step, according to its model card. Bilibili reports scores of 0.8789 on FLORES COMET-22 and 75.35 on its WMT26 Judge scale for the 9B model, and it compares the family against GPT-5.6 Sol, DeepSeek V4.1 Flash and Gemini 3.5 Flash Lite. On those tables the closed models score higher on WMT26 Judge (89.10 for GPT-5.6 Sol against 75.35), while Bilibili reports Index-Translate-9B as top among the systems in its report on instruction following for low-resource languages. Those are Bilibili's own evaluation settings, not an independent test.
The same team maintains IndexTTS, an open zero-shot text-to-speech system that clones a voice from a single reference clip, the same job that hosted tools such as ElevenLabs, Fish Audio and Resemble AI sell as a service. Its GitHub repository shows about 24.9k stars, and the latest release, IndexTTS-2.5, lists Chinese, English, Japanese, Spanish and Arabic with emotion control. Earlier work includes the Index-1.9B series, a lightweight bilingual language model family with Base, Chat and Character variants and 32K context, last updated in June 2024.
For funding, Bilibili is a listed company, so there is no private round history to report. Its investor relations site lists two September 2026 convertible note deals: US$700 million priced on the 4th, and a US$500 million marketed offering completed on the 9th. The company does not break out AI spending on that page, so Hokai does not estimate what the Index team costs.
Where it fits: Index-Translate competes with translation tools such as DeepL and with general models from OpenAI, DeepSeek and Google DeepMind. Its pull is open weights you can run yourself, including GGUF builds for small machines. Its limit is that the 35B-A3B version is a preview and that no trust center or compliance certification for the Index models was found on the pages Hokai read.
Mission
To enrich the everyday lives of young generations in China.
Products
- Index-Translate (Open-weights translation models): https://huggingface.co/collections/IndexTeam/index-translate
- IndexTTS (Open zero-shot text-to-speech system): https://github.com/index-tts/index-tts
- Index-1.9B (Lightweight bilingual language models): https://huggingface.co/collections/IndexTeam/index-19b-series
- Bilibili (Video community platform): https://www.bilibili.com
Bilibili Models on Hokai
Links
Frequently Asked Questions
How is Bilibili funded?
Bilibili is publicly listed (NASDAQ: BILI, HKEX: 9626), so it reports no venture round total. Its investor relations site lists pricing of US$700 million convertible senior notes offerings on 4 September 2026 and completion of a US$500 million marketed offering on 9 September 2026. Hokai found no separate AI budget figure.
What AI products does Bilibili release?
The Index team publishes Index-Translate in 2B, 9B and 35B-A3B preview sizes, the IndexTTS voice-cloning system, and the Index-1.9B chat models. All are free to download from Hugging Face or GitHub, and Index-Translate-9B is licensed Apache-2.0. The video platform itself is a separate consumer product.
Does Bilibili's AI work come with compliance certifications?
Hokai found no trust center, SOC 2 report or ISO certificate for the Index models on the pages it read. Because the weights are downloadable, you can run Index-Translate on your own servers and keep text off Bilibili's systems. Confirm the license on each repository before commercial use.
How does Bilibili's Index-Translate compare with closed translation models?
In Bilibili's own model card, GPT-5.6 Sol scores 89.10 on WMT26 Judge against 75.35 for Index-Translate-9B, so closed models still win that axis. Index-Translate-9B wins on cost control, since you host it, and Bilibili reports it ahead on low-resource instruction following. Pick [DeepL](/hub/tools/deepl) or a closed API for top quality, and Index-Translate when data must stay in-house.
How do you start using Bilibili's Index models?
Open the Index-Translate demo at index-translate.bilibili.com to try it with no setup. For your own stack, download Index-Translate-9B from Hugging Face and serve it with vLLM through an OpenAI-compatible client, or take a GGUF build for a small machine. IndexTTS installs from its GitHub repository.