Qwen3.8-Flash (open weights: Qwen3.8-Flash-Next)
Alibaba (Qwen) · Multimodal model
Qwen3.8-Flash is Alibaba's low-cost Qwen model, served on the API with a 1M-token context. Its open-weight base, Qwen3.8-Flash-Next, is a 125B mixture-of-experts model with only 6B active parameters that accepts images and video.
Current Open weights
ReleasedAug 26, 2026
Context window1M tokens (API default); 262K native for open Flash-Next weights, extensible to 1M
Input → outputtext, image, video → text
API priceqwen3.8-flash (International/Singapore, 0 < tokens ≤ 1M): $0.15 / 1M input tokens, $0.47 / 1M output tokens
Consumer access—
Best for
Cheap, fast Qwen model for high-volume tasks with long context and image input.
Overview
Qwen3.8-Flash is Alibaba's low-cost Qwen model, served on the API with a 1M-token context. Its open-weight base, Qwen3.8-Flash-Next, is a 125B mixture-of-experts model with only 6B active parameters that accepts images and video.
Strengths and limits
- Low API price with a 1M-token context
- 125B total / 6B active MoE, efficient to serve
- Reported SWE-bench Pro 62.5 and GPQA Diamond 91.7
- Open weights use the qwen-community-1.0 license, not Apache 2.0
- Production features (1M default context, built-in tools) are API-only
Published benchmark scores
SWE-bench Pro62.5
Source: huggingface.co
GPQA Diamond91.7
Source: huggingface.co
HLE35.9
Source: huggingface.co
LiveCodeBench v691.9
Source: huggingface.co
Scores are as published by the source; settings (reasoning effort, tools, attempts) differ between labs, so compare with care. How to read benchmarks →