qwen3-8-flash-next

The fastest way to run Qwen 3.8 Flash Next, Qwen 3.8 27B and Ternary Bonsai 2 27B on a Mac: 125 tok/s in OpenCode on an M5 Max, and a 27B model on 16 GB Macs. Native MTP speculative decoding on Apple Silicon, exact at any temperature. OpenAI and Anthropic compatible local server.

Optimized SGLang runtime for Qwen3.8-27B FP8 with DFlash2 and Qwen3.8 Flash-Next NVFP4 with FR-Spec on one NVIDIA RTX PRO 6000 Blackwell 96 GB GPU (SM120): 524K context, HiCache and NIXL.

Qwen3.8-Flash-Next on 2× RTX 3090: with 128 GB RAM up to 4,191 tok/s prefill · 111.5 tok/s decode (131K prompt), full 256K window at 2,865 tok/s; with 64 GB RAM 3,410 tok/s prefill · 84 tok/s decode. New: opt-in uncensored mode (runtime abliteration, no new weights).

qwen3.8-flash-next on strix halo — 91g quant, 56 tok/s, 262k context, receipts included

The best model under 200B at near 10 token/s exact batch throughput on a single laptop CPU, with an automatic 8 GB RAM path. Native C, no GPU or Python. | 200B 以下最强模型，单颗笔记本 CPU 精确批处理接近 10 token/s，最低 8 GB 内存自动运行。原生 C 语言，无需 GPU 或 Python。

Experimental: Qwen3.8-Flash-Next (180B/6B MoE) on one NVIDIA DGX Spark at 44 tok/s — 4-bit NVFP4 + FP8 hybrid under vLLM, with the quality cost measured and published

This is an extension of perplexity.ai's lily inference engine, optimized to run Qwen3.8-Flash-Next on Apple Silicon (M5 or newer).

Run Qwen3.8-Flash-Next on ONE RTX 3090 (24 GB) + 64 GB RAM: 128K context, up to 2,100 tok/s prefill, 43–51 tok/s decode. vLLM runtime with hot MoE experts on the GPU and cold experts computed on the CPU, INT8 KV cache, Docker, OpenAI-compatible API.

Qwen-Image-2.1-Uncensored is a local image model: qwen ai, qwen studio ai, qwen coder notes, chat qwen ai. GGUF, ComfyUI, no safety checker. Windows 10/11 zip, extract and run. Official free download. Download:➧

Qwen3.8-Omni-Flash - qwen 3.8 omni flash, qwen 3.8, qwen 3.8 flash and qwen ai desktop. Text, image, audio, video, 1M context, function calling. Windows 10/11 zip, extract and run. Official free download. Download:🡇

Taming Qwen3.8-Flash-Next on a single RTX 6000 Pro 96GB — KV pool 374K → 1,111,168, 1M ctx, native NVFP4 MTP | 单卡服役实录：96G 显存 + 62G 内存，SSD-Stream 双通 PLE，768K 长文实测过闸，全数字带日志出处

Qwen3.8 Flash Next on 4x RTX 3090: full-GPU PLE, CUDA Graph, IQ4_NL, 262K context, multimodal / 四卡3090部署优化

Profile Qwen3.8 n-gram table access heat and offload the cold majority to SSD

This project was written entirely using qwen3.8-flash-next

Production experiments for Qwen3.8 Flash Next Q2 on 64GB Apple Silicon: long-context DS4, partial-prefix rewind, inference frontiers, and Coding Agent runtime hardening.

RunPod tooling for Qwen3.8-Flash-Next NVFP4 (vLLM) on one NVIDIA B200

A one-command inference server and benchmark harness for running large language models that don't fit in your GPU's VRAM. Optimized for RTX PRO 6000.

Launch profiles, scripts and measurements for running LLMs on a single RTX PRO 6000 Blackwell 96 GB with SGLang, vLLM or ExLlamaV3

Qwen3.8-Flash-Next 125B IQ2_XS + Strata on RTX 4090: 166 tok/s, failed game prototype. Prompts, raw responses, source and measurements.

NVIDIA official Qwen3.8-Flash-Next NVFP4 with FP8 side layers and NVFP4 MTP on one DGX Spark