This model was retired on 30 October 2026 and no longer accepts API requests. Recommended replacement: DeepSeek V4 Flash 0731.
Qwen3.5 4B
Retired
A compact 4B parameter reasoning model with 262K context — comparable to GPT-OSS-20B at a fraction of the cost.
Total Parameters
4B
Context Window
262K
Native
Intelligence
13.1
AA Index v4.3
Provider
Alibaba
Cloud
Big Reasoning in a Small Package
Qwen3.5-4B is a compact 4 billion parameter reasoning model from Alibaba Cloud's Qwen family, featuring a native 262K token context length. Despite its small size, it delivers remarkably strong performance on complex reasoning tasks — Qwen's benchmarks show it is comparable to the much larger GPT-OSS-20B model. This makes it an ideal choice for cost-sensitive workloads that still require robust intelligence.
💡 Developer Tip
At just $0.04/$0.06 per 1M tokens on Standard tier, Qwen3.5-4B is one of the most cost-efficient reasoning models available — perfect for high-volume batch workloads.
Compact Reasoning — 4B parameters
Built for efficient reasoning at scale
Compact Reasoning Powerhouse
Despite its small 4B parameter size, Qwen3.5-4B delivers reasoning performance comparable to GPT-OSS-20B on complex tasks — at a fraction of the cost.
Cost-Efficient Agent Workflows
Ideal for high-volume agentic pipelines where per-token cost matters. Run thousands of concurrent reasoning tasks without breaking the bank.
Edge & Lightweight Deployment
Small enough for resource-constrained environments while maintaining strong reasoning capabilities. Perfect for on-device or low-latency scenarios.
Long-Context Processing
Native 262K token context window enables processing of large documents, codebases, and multi-turn conversations without chunking overhead.
Punches Above Its Weight
Comparable to GPT-OSS-20B on complex reasoning tasks, at 5x fewer parameters.
Intelligence Index
Better than 71% of models
GPQA Diamond
Better than 75% of models
τ²-Bench Telecom
Better than 92% of models
| Category | Benchmark | Score | Description |
|---|---|---|---|
| Reasoning | GPQA Diamond | 77% | Graduate-level scientific reasoning |
| Reasoning | τ²-Bench Telecom | 92% | AI agents in dual-control scenarios |
| Reasoning | AA-LCR | 56% | Long context reasoning evaluation |
| Reasoning | IFBench | 52% | Instruction-following accuracy |
| Reasoning | GDPval-AA | 27.1% | Agentic performance on real-world work tasks |
| Coding | SciCode | 16% | Python for scientific computing |
| Coding | Terminal-Bench Hard | 18% | Agentic coding & terminal use |
| Knowledge | AA-Omniscience Accuracy | 13% | Proportion of correctly answered questions |
| Knowledge | AA-Omniscience Non-Hallucination | 20% | Confidently answered questions that are correct |
Metrics sourced from Artificial Analysis.