Tencent

A Qwen3-Embedding fine-tune for skill retrieval in agent harnesses.

A 295B-A21B MoE from Tencent. It improves over its predecessor across all metrics. Most notable, however, is the license change: While the previous version (covered in Artifacts 21) used a custom and rather restrictive license, Tencent switched to Apache 2 for this release. The model was also able to proof a 50 year old math problem (with a dedicated harness and Sol as a judge, although it is unclear how important the latter really is).

Another open LLM by Tencent with its prohibitive license, which completely forbids use in the UK, South Korea, and the entire EU.

A very capable image generation model with a whopping 80B (total) parameters, 13B active. Just like other Tencent models, the license is one of the most restrictive ones out there.

A small VLM. Similar to other Tencent models, the license is rather restrictive, banning usage in the EU outright.

Another world/video generation model by Tencent.

An update to Tencent's video generation model. Just like their other models, it comes with a very restrictive license.

Another 3D generation model by Tencent. Just like their other models, the license is one of the most restrictive ones out there.

A very capable image generation model, released, just like the other Tencent models, under a restrictive license.

A specialized translation model covering a lot of Chinese ethnic minority languages. It is released under the restrictive Hunyuan license.

A video game generation model, which you can control with a keyboard and mouse. Similar to other models by Tencent, it comes with a very restrictive license.

Dense versions of the MoE models we featured last episode. However, they share the same bad license, disallowing the usage (and sharing outputs!) of the models in the EU, UK and SK.

While Genie 3 rightfully made headlines and is unquestionably leagues ahead of the competition, others are catching up (or at least participating).

Tencent releases both a 7B dense and a 80B total / 13B active MoE model. It also features 256K context, has very solid benchmark scores (including function calling). Aside from that, people actually use it and are impressed by it! We sound like a broken record, but Chinese labs and companies continue to outbid each other in the open model space with very solid models. Edit: Unfortunately this model is up there in terms of bad licenses. A portion of it states: "You must not use, reproduce, modify, distribute, or display the Tencent Hunyuan Works, Output or results of the Tencent Hunyuan Works outside the Territory. Any such use outside the Territory is unlicensed and unauthorized under this Agreement." Here, the territory is everywhere but the EU, UK, and SK.

An update to the Hunyuan3D model.

A version of the video generation model HunyuanVideo which supports images as inputs.