Qwen3.8-Flash-Next on Glyd

A 180-billion-parameter model on one gaming GPU.

Qwen3.8-Flash-Next runs here on this computer's GPU with Glyd's own engine and compression. Nothing you type leaves this computer.

Fast and precise

This is Glyd's Hummingbird level, its most compact. Fast mode keeps every expert in Glyd's smallest format, so more of them fit on the GPU. Precise mode adds the precision pack: more bits where they matter, and stays closer to the original model. In both, the engine guesses the next word with the model's own prediction head and checks it in the same pass: faster answers, the same words.

Use it from your apps

• OpenAI-compatible: , model qwen3.8-flash-next, any key
• Anthropic-compatible:
• Streaming, tool calls and reasoning work through both. Follow-up messages reuse the conversation, so they answer in a fraction of a second.

Source and license

Scripts, benchmarks and results: github.com/surya-koritala/glyd-qwen-3.8-flash (Apache-2.0). The engine is under the Business Source License 1.1: free for personal and non-commercial use.