unsloth/qwen3.6-27b
Qwen3.6-27B (unsloth)
Throughput
9.4t/s@ 64k
Engine
vLLM
Parameters
27B
Released
2026-04-22
Benchmarked
2026-06-27
Context ladder
Throughput at each benched context window (single measurement).
| Context | KV | Throughput |
|---|---|---|
| @ 64k peak golden | — | 9.4t/s |
Golden profile
unsloth-qwen3-6-27b-eugr
Capabilities
generaldensevllmllamacppnvfp4ggufagentic
Why we run it
Dense 27B Qwen3.6 line for Spark bake-off — NVFP4 (eugr vLLM), GGUF (llama.cpp), and MTP GGUF variants.
Bench notes
golden ?/? @ 9.2 tok/s — fill~50000 — bench-agent-v2 — tool_ok=False