LLMs8 min read
Qwen3.8 27B Model Adds Large Numbers in Words
A test showed Qwen3.8-27B computes sums and returns the answer in English words. It succeeded when reasoning was enabled but failed without it.
From Simon Willison on LLMs
Loading
90 days of the enterprise, pro and team plans free, no card needed. See the plans
Blog
Daily notes on new models, LLM releases, agent frameworks and AI research, written from the sources we follow and delivered as a newsletter every day.
Get the morning and evening issues
Every new post of the day, in one email. Confirmation required.
LLMs8 min read
A test showed Qwen3.8-27B computes sums and returns the answer in English words. It succeeded when reasoning was enabled but failed without it.
From Simon Willison on LLMs
AI1 min read
Anthropic has launched a new evaluation process for Claude Code plugins, allowing developers to measure skill triggering, survival across edits, and performance against a bare model. This workflow uses six grader types, costing $0.41 per run and providing detailed insights into plugin effectiveness.
From MarkTechPost
Photo: Yan Krukau
LLMs1 min read
Astra produces better visual outputs than Sol at similar or lower token costs, with lower input token counts and improved image quality across reasoning levels.
From Simon Willison
LLMs1 min read
Google released Gemini 3.8 Flash, offering low, medium, and high thinking levels, with improvements in speed, cost, and HTML/JavaScript capabilities for developers.
From Simon Willison
LLMs1 min read
Claude Fable 5.1 achieves a 52.6% score on the Terminal-Bench-Science 0.1 benchmark and features five reasoning levels, affecting task performance and output length.
From Simon Willison
Research1 min read
Microsoft Research introduced GigaPath-Flash and GigaTIME-Flash, which lower computational requirements while maintaining performance, enabling larger-scale pathology studies.
From Microsoft Research
Research1 min read
DeepMind has piloted the first double-blind evaluation process for AI systems, aiming to improve assessment objectivity and reliability in model performance testing.
From Google DeepMind blog
LLMs1 min read
A new 4-bit model, dubbed Quantization-Aware Healing, achieves performance comparable to its full-precision original. This technique offers a compressed model size with minimal impact on accuracy for running AI agents.
From Hugging Face blog
LLMs1 min read
Hugging Face introduced the ASRA benchmark to evaluate and compare the optimization of speech recognition models. The benchmark focuses on measuring the impact of various optimization techniques on model performance.
From Hugging Face blog
LLMs1 min read
Together AI allows engineers to conduct A/B tests in production by splitting traffic between model variants. This enables real-world measurement of model performance against the control, using a flexible endpoint-level routing system.
From Together AI blog
Research1 min read
Gemini 3.7 Flash is a new language model introduced by Google DeepMind, designed for improved performance and efficiency in AI applications.
From Google DeepMind blog
LLMs1 min read
Hugging Face replicated 2,200 research papers from ICML, providing accessible implementations and datasets. This effort offers engineers a resource for understanding and evaluating model performance directly.
From Hugging Face blog
Models1 min read
Google AI revealed new developments in July 2026, including model updates and tools that impact model deployment and management.
From Google AI blog
Posts are drafted from public feeds by models OpenSmartRoute routes to - the same router, skill and metering customers use - and always link to the original source. Corrections: support.
Archive (100)