Tencent Hy4

Tencent's Hy4 is 770B parameters with 49B active and a 1M context. We diffed the licence against Apache 2.0: identical through Section 9, with one MaaS clause added. And Hy4-flash has a 63.4 GB Engram floor that no quantisation touches.

Haru-4B

Haru-4B is the most-liked open model of the month and almost nobody has run it. Apache 2.0 full-duplex speech from a Japanese lab: three streams summed into one LM position, continuous latents with a one-step MeanFlow head, 40M hours of pretraining, and no published benchmarks at all.

Kling Image 4.0

Kuaishou's Kling Image 4.0 ships native 8K and 2,000 characters of in-image text, with no weights and no readable API docs. The full schema is in a ComfyUI node merged the night of the launch, reference images are now billed, and the seed is decorative.

Kimi K3.1 Pro

Moonshot published the Kimi K3.1 Pro weights ungated on 2 November. The reported 90.5 to 81.0 cyber cut is real, but K3 scored 33.3 on the same test and DeepSeek-V4.1-Flash scores 85.7 under MIT. What 2.8T at 4 bits actually buys you.

Edge0-145B

Edge0-145B is not a Qwen requant: the routed MoE experts were replaced with a latent MoE and retrained at 4 bits. 30% fewer expert parameters, double the decode on a 16 GiB budget, and a commercial coding assistant needs permission from both Edge0 and Qwen.

Nemotron 3.5 Bolt E4B

NVIDIA Nemotron 3.5 Bolt E4B is post-trained from Google Gemma 4 E4B. The configs differ in one field: the context window is halved. What that buys you, and where Qwen 3.5 4B still wins.