AI Weekly Malaysia

LiquidAI released QAD (Quantization-Aware Distillation) Q4_0 GGUF checkpoints for four LFM2.5 models (230M, 350M, 1.2B-Instruct, 2.6B), recovering ~97% of the BF16 accuracy typically lost to 4-bit quantization while maintaining Q4_0 memory footprint and throughput. Benchmarks across GPQA Diamond, MMLU-Pro, IFEval, BFCLv4, and others show the QAD checkpoints match or exceed Q5_K_M and Q4_K_M quality at 3-33% higher decode throughput, tested on MacBook Pro, NucBox EVO-X2, Samsung Galaxy S26 Ultra, and Raspberry Pi 5.

Why: If you're running LFM2.5 models on edge hardware via llama.cpp, swap your existing PTQ Q4_0 or Q4_K_M GGUFs for these QAD Q4_0 files to get measurably better reasoning, instruction-following, and tool-use accuracy at the same memory and speed — the 1.2B and 2.6B checkpoints are the most relevant for agentic workloads.

Inherent, a London AI lab founded by Google DeepMind alumni that recently raised a $50M seed round, claims its AI agent Faraday outperformed Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.5 at independently reproducing findings from published scientific papers. Notably, Faraday runs on Qwen 3.6, a 27-billion-parameter model—far smaller than the frontier systems it claims to beat.

Why: If the benchmark holds up under scrutiny, it suggests that for narrow agentic tasks like research replication, a well-prompted 27B model can match or beat frontier-scale models—meaning builders may not need to pay for the most expensive API calls for certain agent workflows. However, this is a self-reported result from a startup promoting its own product, so verify against independent replication before changing your model selection.