Basalt inference engine launched with up to 2.6x speedup

Developers released Basalt, an MIT-licensed inference engine for Blackwell GPUs, tuned for Qwen3.8 Flash-Next. It claims up to 2.6x Strata's throughput, supports dual GPUs and eight concurrent streams.

• Qwen3.8-Flash-Next gains Strix Halo support on Strata2026-10-07
• User Shares Strata Speedup for Local LLMs but 64GB RAM Constraint2026-10-04
• Swift1.5-Qwen3.8-Flash-Next shines in benchmarks2026-09-25
• Qwen3.8-Flash-Next Hits 65 Tokens/Second on 12GB VRAM2026-09-24
• Qwen 3.6 35B praised for general agentic tasks2026-10-09