Qwen3.5 27B

Qwen3.5 27B is a native vision-language model from the Qwen team, released under open weights so that developers can run, fine-tune, and self-host it. Its architecture is described as a dense model that incorporates a linear attention mechanism, which is intended to deliver fast response times while balancing inference speed and overall performance. According to OpenRouter's listing, Qwen3.5 27B reports overall capabilities that are comparable to those of the larger Qwen3.5-122B-A10B, suggesting it sits in a similar capability band despite its smaller active footprint.

The model is positioned for multimodal workloads that benefit from a unified vision-language design rather than a separately vision-tuned variant. Early fusion training on multimodal tokens lets it handle text, image, and video inputs in a single context, and the inference profile has attracted active community optimization work, including an NVIDIA developer forum thread focused on squeezing higher tokens-per-second throughput from the open weights on DGX Spark hardware. Practical fit for Qwen3.5 27B includes image and video understanding alongside general text reasoning, making it attractive for teams who want multimodal coverage with the cost and latency profile of a dense 27B-scale model.

Jalapeno CloudQwen3.5-27Bqwen

Open provider page Provider docs

Quick Info

Provider: Jalapeno Cloud
Model key: Qwen3.5-27B
Release date: Feb 23, 2026
Last updated: Feb 23, 2026
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $0.30
Output token cost: $2.40

Limits

Output tokens: 65,536 tokens
Context window: 262,144 tokens

Transparent token rates

Compare Qwen3.5 27B pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about Qwen3.5 27B

Jalapeno Cloud

Coverage

The Ollama library listing for qwen3.5:27b documents the exact Qwen3.5-27B variant as a 27.8-billion-parameter multimodal model released under Apache License 2.0, with vision, tools, and "thinking" mode supported. The page attributes the model to the Qwen team's Qwen 3.5 family and notes the upstream architecture combi The same listing surfaces operational details relevant to developers self-hosting the exact Qwen3.5-27B subject: a Q4_K_M quantization shipping at roughly 17 GB, a model architecture tag of "qwen35," and an updated release timestamp of approximately six months before the current date. Recommended sampling defaults show

Jalapeno Cloud

CoverageBenchmark

OpenRouter describes Qwen3.5-27B as a native vision-language dense model using linear attention, with input and output modalities, a 262K context window, and a Feb. 25, 2026 release date. The page also reports $0.195 per million input tokens and $1.56 per million output tokens for Alibaba Cloud International, while lis For model comparison, the page lists six providers and reports a weighted average input price of $0.2396 per million tokens and output price of $2.178 per million tokens. It reports 32 tokens per second as the best P50 throughput and 0.29 seconds as the best P50 latency across providers; these figures are cross-provide