v2026.05

New models

• Gemma-4 (known issue #128)
• InternVL3.5

Other updates

• Video input implemented for all models that support it.
• q4_k quantization works now.
• A modified version of WebUI from llama.cpp is used.

v0.22

New Models

• Qianfan-OCR
• Penguin-VL

Bug fixing

• A nearly one-year old bug (Issue #65)

v0.21

New models:

• Qwen3.5
• GLM-OCR
• Youtu-VL
• Youtu-LLM
• Qwen3-TTS (voice clone only support xvec)

v0.20

Support new models:

• QWen3-ASR
• QWen3-ForcedAligner

v0.19

• As always, more models are supported. Note that most of the new models are special in some way.

  • Step3-VL: strong vision capability
  • GLM-4.7-Flash: strong coding capability
  • TranslateGemma: translation
  • WeDLM: diffusion with AR
  • QWen3-VL-Embedding/Reranker: multimodal embedding
  • HY-MT: translation
  • GLM-ASR-Nano: ASR
  • Qwen3-VL: strong vision capability

v0.18

• As always, more models are supported.
• Windows: prebuilt binary with Vulkan (1.4.335.0). Use -ngl all to run whole model on default GPU.
• New server.exe with built-in llama.cpp WebUI

v0.17

• As always, more models are supported, notably LLaDA2.0.
• Windows: prebuilt binary with Vulkan (1.4.321.1). Use -ngl all to run whole model on default GPU.

v0.16

• As always, more models are supported, notably Janus-Pro.
• Windows: prebuilt binary with Vulkan (1.4.321.1). Use -ngl all to run whole model on default GPU.

v0.15

• As always, more models are supported.
• Windows: prebuilt binary with Vulkan (1.4.321.1). Use -ngl all to run whole model on default GPU.

v0.14

• Fix main_nim.exe: could not download models that are > 2GB due to this.