dasllama.io newsdasllama.io2026-08-18T00:00:00Zhttps://dasllama.io/Qwen 3.8 27B joins the board - the new dense hybrid runs on CPU and Metal.https://dasllama.io/#n-2026-08-18-qwen38-27b2026-08-18T00:00:00Z<p>Qwen 3.8 27B is out, and it needed no new architecture support in dasLLAMA: the 3.8
generation keeps the qwen3.5 architecture - Gated-DeltaNet recurrent layers interleaved
3:1 with gated full attention - so it loads on the existing family path as a pure
scale-up. The Q4_K_M file's attention and FFN planes serve natively on the kq rails, CPU
and Metal alike (the DeltaNet projections transcode to q8 at load), and the model joins
the official board catalog.</p>
<p>On an M1 Max (pp512 / tg128, llama-bench protocol): Metal - das 147.0 / 14.5 tok/s,
llama.cpp 137.6 / 11.5; CPU - das 33.2 / 5.8, llama.cpp 29.8 / 5.6. On an M4 Pro:
Metal - das 127.0 / 12.8, llama.cpp 126.2 / 11.5; CPU - das 45.8 / 11.6, llama.cpp
40.4 / 11.0. Full rows for both boxes are on the
<a href="https://daslang.io/dasllama.html">board</a>.</p>Metal support lands in speech-to-text - whisper large-v3-turbo at 49x realtime on an M1 Max.https://dasllama.io/#n-2026-08-17-metal-asr2026-08-17T00:00:00Z<p>Speech-to-text now serves on Metal: whisper's cross-attention and decoder run on the GPU
by default on Apple silicon, and the full-GPU mode powers the new gpu category on the
board. On an M1 Max, whisper large-v3-turbo transcribes a 3:19 clip in 4.0 s - 49x
realtime, against 36x for whisper.cpp's Metal path and 12x for our CPU lane. The
<a href="https://daslang.io/dasllama.html">board</a> now splits ASR into the same cpu / cpu + accel /
gpu categories the LLM board has.</p>dasllama.io is live - news, the full ladder, and the sidecar exchange.https://dasllama.io/#n-2026-08-10-dasllama-io-is-live2026-08-10T00:00:00Z<p>Everything measured now lives here, community submissions included.
<a href="https://daslang.io/dasllama.html">daslang.io/dasllama.html</a> stays the promotional
scoreboard and carries official measurements only.</p>