Matching stories Newest first
Eagle-4 speculative decoding method reported faster than DFlash2 and DSpark on Qwen3.8-27B
80x. cpp and vLLM support later.
Qev-0.8b tiny local decision model claimed as new state of the art
8b as a new state-of-the-art tiny local decision model. The post itself is a brief community submission with no benchmark details in the excerpt.
Intel Arc Pro B65 four-GPU rig posts long-context throughput sweeps for Qwen3.8 models
8 Flash-Next and 27B models on 1x to 4x Intel B65 setups with vLLM, reporting output throughput across context lengths up to 200K. Results include queueing effects and are not model-quality scores.
Six-way benchmark compares Qwen3.8-Flash-Next locally against Claude Opus 5.5 on a Rust feature
A community test of 12 PRs on a LlamaStash feature found Opus 5.5 averaged 82% against 68% for Claude Code with Flash-Next and 66% for Pi with Flash-Next. Flash-Next ran locally at about 4.16 bpw, so the author paid only electricity, and results are a single developer's unverified experiment.
Vals AI says DeepSeek V4.1 Flash 1017 beats GPT-6 Sol on a multi-step offensive cybersecurity task
Vals AI reports that DeepSeek V4.1 Flash 1017 outperformed GPT-6 Sol on its multi-step offensive cybersecurity task. Sol was tested through OpenAI's Daybreak trusted access program.