Underdog releases Saluki 27B, a 2-bit Qwen3.8-27B under Apache 2.0
Overview
Underdog, the on-device assistant from Conway Research, has released Saluki 27B under Apache 2.0, a 2-bit GGUF of Qwen3.8-27B that fits in 7.89 GB, compared with 54 GB for the full BF16 model.
According to MarkTechPost's report, Saluki scores 88 against 84 for the full model on Underdog Bench and raises parallel tool-call accuracy to 42 from 35. It runs on stock llama.cpp.
The same report says math and reasoning drop sharply, with AIME 2025 at 79.2 versus 96.7 for the original, so the tool-calling gains come with weaker performance on those tasks.
Written by AI from the articles below · updated Oct 9, 4:53 AM ET
Check the sources:
Article timeline
Follow the coverage from different perspectives. Times are ET.
- MarkTechPostUnderdog Releases Saluki 27B, a 2-Bit Qwen3.8-27B That Beats the Original at Tool Calling
AIUnderdog has released Saluki 27B under Apache 2.0, a 2-bit GGUF of Qwen3.8-27B that fits in 7.89 GB, versus 54 GB for the full BF16 model. On Underdog Bench, Saluki scores 88 against 84 for the full model, and it raises parallel tool-call accuracy to 42 from 35. It runs on stock llama.cpp, but math and reasoning drop sharply, with AIME 2025 at 79.2 versus 96.7.
Heat trend
Not enough continuous observations to show a trend yet.