BLAST RADIUS

Open-weight & owned stack · Quantization efficiency

FreeToken fork adds DeepSeek vision and decoding

Sep 22, 2026

A community FreeToken fork added DeepSeek-V4.1-Flash, vision, speculative decoding and tensor parallelism with reported speeds up to 48 tokens per second on two RTX 3090 GPUs.

Read the original at reddit.comOpens the publisher's site in a new tab

More in Open-weight & owned stack