Qwen 3.5 9B
A 9-billion-parameter language model in Alibaba's Qwen (Tongyi Qianwen) series, supporting local deployment and designed for scenarios with consumer-grade VRAM constraints.
Core Facts
Timeline (last 90 days)
A user running the Qwen 3.5 9B model via Ollama on EndeavourOS (Arch Linux) encountered a problem where the model, after outputting its thinking process, produced no final answer.
12GB of VRAM is sufficient for running a 9B model; the problem is more likely at the software level than the hardware level.
Moss Image uses Qwen 3.5 9B, whose parameter count is far larger than the diffusion model itself, as its text encoder.
In MOS Image's ComfyUI configuration, Qwen 3.5 9B can be selected as the text encoder and Flux 2.0 VAE as the VAE.
PewDiePie's Ajax project is a fine-tuned Qwen 3.5 9B model; the reason for fine-tuning was that after switching to the local Qwen 9B, the agent's tool calls often failed or the wrong tool was selected.
On Bespoke Labs' leakage-level evaluation, Nimble reaches 90.12, while its base model Qwen 3.5 9B scores only 66.36.
The author subjectively rated Bonsai's pagoda scene as having the richest detail and leading quality, with Gemma 4 coming second, and Qwen 3.5 9B producing broken output.
Qwen 3.5 9B (Q6_K) output at approximately 168 tok/s, faster than Bonsai 2 27B's 101 tok/s.
Qwen 3.5 9B is a mid-parameter version of Alibaba's Tongyi Qianwen, and its 9B parameter count allows it to run on a single consumer-grade GPU.
CrofAI's so-called proprietary greg series models were actually routed to other models: greg-2-ultra routed to GLM 5.2, greg-1-mini routed to Qwen 3.5 9B, and greg-2-super/greg-1/greg-1-super routed to Kimi K2.7 Code.
3 more timeline events
All Facts (11)
The Qwen2.5:7B model is approximately 4 GB in size after Q4 quantization.
80%Unverified12GB of VRAM is sufficient for running a 9B model; the problem is more likely at the software level than the hardware level.
50%UnverifiedMoss Image uses Qwen 3.5 9B, whose parameter count is far larger than the diffusion model itself, as its text encoder.
50%UnverifiedIn MOS Image's ComfyUI configuration, Qwen 3.5 9B can be selected as the text encoder and Flux 2.0 VAE as the VAE.
50%UnverifiedOn Bespoke Labs' leakage-level evaluation, Nimble reaches 90.12, while its base model Qwen 3.5 9B scores only 66.36.
50%UnverifiedQwen 3.5 9B is a mid-parameter version of Alibaba's Tongyi Qianwen, and its 9B parameter count allows it to run on a single consumer-grade GPU.
50%UnverifiedCrofAI's so-called proprietary greg series models were actually routed to other models: greg-2-ultra routed to GLM 5.2, greg-1-mini routed to Qwen 3.5 9B, and greg-2-super/greg-1/greg-1-super routed to Kimi K2.7 Code.
50%UnverifiedWhen writing scripts, Qwen 3.5 9B provides thoughtful features such as backups and debug output but often comes with 1–3 errors, whereas ActionAssist tends to generate concise and accurate code with a lower error rate.
50%UnverifiedA user running the Qwen 3.5 9B model via Ollama on EndeavourOS (Arch Linux) encountered a problem where the model, after outputting its thinking process, produced no final answer.
50%UnverifiedPewDiePie's Ajax project is a fine-tuned Qwen 3.5 9B model; the reason for fine-tuning was that after switching to the local Qwen 9B, the agent's tool calls often failed or the wrong tool was selected.
50%UnverifiedA fine-tuned version based on Qwen 3.5 9B called Emperios 3.8 distill has appeared in the community.
50%