Qwen3.8 27B and Muse Glimmer Benchmarks: Accuracy, Token Efficiency and Memory Use
Qwen3.8 is remarkably capable, while Muse Glimmer reveals a very different approach to reasoning and memory efficiency.
The Kaitchup – AI on a Budget
Weekly tutorials and news on adapting large language models (LLMs) to your tasks and hardware using the most recent techniques and models. The Kaitchup proposes a collection of 180+ AI notebooks regularly updated.
Live Last read · last published · next check
Latest posts
Qwen3.8 27B, Nemotron 3.5, Muse, DeepSeek V4 Pro: A Huge Week for Open-Weight AI
The Weekly Kaitchup #155
Laguna S 2.1: How Agent Harnesses and Inference Budgets Shape Coding Performance
Testing long-horizon coding performance on DeepSWE and Terminal-Bench 2.1
Muse Glimmer: Meta’s 30B Model Built for Efficient Inference
Inside Meta’s 30B local reasoning model and its tiny KV cache
Qwen3.8 Is Almost Here — and Agent Benchmarks Are More Fragile Than They Look
The Weekly Kaitchup #154
ThinkingCap-Qwen3.6-27B Review: 2x Fewer Tokens, Same Accuracy?
A faster, more stable Qwen3.6 for local AI inference
DeepSeek-V4-Flash-0731 and Inkling Small: Smaller, but Better?
The Weekly Kaitchup #153
Bonsai 27B Review: Can a 3.9 GB 1-Bit Model Match Qwen3.6 27B?
An in-depth look at Bonsai 27B’s accuracy, token efficiency, reasoning stability, and production trade-offs.
Agentic AI at Two Different Scales: Nanbeige4.2-3B and Laguna S2.1
The Weekly Kaitchup #152
Qwen3.8: What Hardware Will You Need to Run Alibaba’s 2.4T Model?
Estimating the memory, storage, and GPU requirements for BF16, NVFP4, Q4, and TQ1 versions of Qwen3.8.
Inkling, Gemma 4 Updates, and 1-Bit Qwen3.6
The Weekly Kaitchup #151