Qwen3.5-Plus

// Overview

Qwen3.5-Plus is the hosted, API-only Plus model in Alibaba's Qwen3.5 generation, launched on 16 February 2026 alongside the open-weight Qwen3.5 family. It sits below the flagship Qwen3-Max as the value tier: cheaper to run while keeping the generation's strongest capabilities. Unlike the downloadable Qwen3.5 weights, Plus is proprietary — you reach it only through Alibaba Cloud Model Studio (Bailian) and resellers such as OpenRouter.

Qwen3.5-Plus is natively multimodal: it accepts text, images, and short video clips as input and returns text. Alibaba's documentation says its multimodal understanding meaningfully outperforms the earlier Qwen3-VL series, while its plain-text quality is comparable to Qwen3-Max. The model carries a 1-million-token context window and supports both a fast 'non-thinking' mode and a chain-of-thought 'thinking' mode you toggle per request.

Qwen3.5-Plus is positioned as an agent model: it is tuned for tool calling and multi-step workflows, and the Plus tier emphasises managed infrastructure and adaptive tool use over raw self-hosting. It was the starting point of the Qwen-Plus line that continued with Qwen3.6-Plus (April 2026) and the GUI-driving Qwen3.7-Plus (June 2026).

// Benchmarks

1. AIME 2026 (math)91.3%
2. GPQA Diamond88.4%
3. MMLU-Pro87.8%
4. MMMU (multimodal)85%
5. LiveCodeBench v683.6%
6. SWE-bench Verified76.4%

Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.

// Pricing

International (Model Studio) rate for the 0–256K input tier; input rises to $0.50/1M above 256K. Includes 1M free tokens for 90 days after activation.

Pricing source ↗

// Strengths

• 1-million-token context window for long documents, codebases, and multi-turn agent traces
• Native multimodal input — text, images, and short video clips — at a low price point
• Toggleable thinking / non-thinking modes, so you pay for reasoning only when a task needs it
• Tuned for agentic tool calling and multi-step workflows
• Inexpensive: $0.40 per 1M input tokens and $2.40 per 1M output tokens on Model Studio

// Best for

• High-volume document and PDF understanding across very long contexts
• Multimodal apps that need to read screenshots, charts, or short video clips
• Tool-using agents and assistants where cost-per-call matters
• Coding and productivity workflows that benefit from a large context but a modest budget
• Drop-in OpenAI-compatible API usage via Model Studio or OpenRouter

// How to access

// Qwen-Plus (multimodal agent) — every version

The full lineage of the Qwen-Plus (multimodal agent) line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.