Qwen3.5 4B

Within the Qwen open-weight family, Qwen3.5 4B is positioned as a compact dense text model. A third-party runner-compatibility index reports it with 4 billion parameters in a dense architecture and links to a Hugging Face repository named Qwen3.5-4B-Instruct, alongside an Ollama tag of qwen3.5:4b, indicating availability for local inference through common open-source runtimes. The same index flags coding and general chat use cases for the variant, framing it as a balanced small-footprint option in the Qwen lineup rather than a frontier-scale system.

On the broader Qwen research index, a vision-language foundation model for autonomous driving called Qwen-Drive-1.0 is described as built on top of the Qwen3.5-4B base, attaching external perception and planning modules to the otherwise untouched multimodal backbone. That reference confirms Qwen3.5-4B as the natively multimodal foundation underlying further Qwen research directions, suggesting the variant is intended not only for direct chat and code workloads but also as a substrate for downstream multimodal extensions.

QVACqwen3.5-4bqwen

Open provider page Provider docs

Quick Info

Provider: QVAC
Model key: qwen3.5-4b
Release date: Nov 1, 2025
Last updated: Nov 1, 2025
Input modalities
Output modalities
Capabilities

Cost

A provider subscription or plan supersedes token-based pricing for this model.

Limits

Output tokens: 8,192 tokens
Context window: 32,768 tokens

Latest news about Qwen3.5 4B

QVAC

CoverageRelease Notes

According to Releasebot's Qwen release-notes feed (last updated Sep 4, 2026), Alibaba's Qwen team introduced Qwen-Drive-1.0, described as the first vision-language foundation model for autonomous driving that unifies 3D perception and visual question answering at the pretraining stage and extends to motion planning. Th The same Qwen release-notes feed (Sep 2, 2026 entry) frames Qwen3.5-4B as natively multimodal at the foundation level, a property leveraged by Qwen-Drive-1.0 without modifying the pretrained VLM architecture. While the feed is a third-party aggregator rather than a Qwen/Alibaba first-party channel, the excerpt reproduc

QVAC

Coverage

Alibaba Cloud's Qwen team (Tongyi Lab) released the Qwen3.5 Small series, including Qwen3.5-4B, alongside 0.8B, 2B, and 9B siblings, with all four models positioned as smartphone-runnable. The Qwen team described Qwen3.5-4B as "a surprisingly powerful multimodal base for lightweight agents," highlighting its suitabilit The Qwen3.5 Small series shares the same native multimodal foundation as the broader Qwen3.5 line, combining a Gated Delta Network with a Mixture of Experts (MoE) mechanism and extended reinforcement learning training. Benchmark comparisons in the report position Qwen3.5-4B (light purple) and Qwen3.5-9B (purple) agains