Newer Models, Same Advantage

Cheaper, Better, Faster, Stronger

Sharing the latest Model Spec

Qwen 3.8 27B is excellent, but it defaults to wildly overthinking things

Friday's big release was Qwen 3. 8 27B , an Apache 2 licensed 27B parameter vision-capable LLM from Alibaba's Qwen research lab.

Introducing the Model Spec

What We Observe as LLM Behavior Can Be a Side-effect of Inference Backend

arXiv:2608. 04714v1 Announce Type: cross Abstract: Benchmark scores are reported as properties of a model, yet the inference framework used to produce them, such as HuggingFace, vLLM, or Ollama, are considered non-influential and their names and versions are almost never disclosed.

Optimizing for the decision not the prediction: an exploration of Smooth Net Benefit as a training objective

The study introduces Smooth Net Benefit (σNB), a differentiable approximation of Net Benefit, as a training objective aimed at aligning predictive models with threshold‑specific clinical decisions. Experiments on the Framingham cardiovascular risk dataset and 44 TabZilla datasets show that σNB training yields modest improvements for logistic regression but little to no benefit for more flexible models such as GAMs and XGBoost. The authors conclude that σNB is not a universal replacement for negative log‑likelihood training, though it may be worth exploring in contexts where model flexibility is limited.

Brevity is the Soul of Inference Efficiency: Inducing Concision in VLMs via Data Curation

Inference efficiency is typically pursued by shrinking the model: distillation, pruning, quantization, and sparse routing each lower per-token cost while treating token count as fixed. But output length has been inflating, and it is precisely the component the standard toolkit leaves untouched.

Mistral 7B

The best 7B model to date, Apache 2.

Brevity is the Soul of Inference Efficiency: Inducing Concision in VLMs via Data Curation

arXiv:2606. 25432v1 Announce Type: new Abstract: Inference efficiency is typically pursued by shrinking the model: distillation, pruning, quantization, and sparse routing each lower per-token cost while treating token count as fixed.

Depth and Scale in the Sub-150M Regime: JugnuLM-53M vs JugnuLM-110M

arXiv:2609.14715v1 Announce Type: new Abstract: We scale our conventional sub-150M pretraining recipe from 53.5M to 109.7M parameters, holding the method fixed (Qwen3-style decoder with grouped-query...

Model Cards

Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and a new price war

The article reports the release of new AI models: Claude Opus 5.5 by Anthropic and GPT‑6 Sol and GPT‑6 Luna by OpenAI, noting that GPT‑6 variants are priced at half the cost of their GPT‑5.6 counterparts. It provides a detailed pricing table comparing input, cached input, and output costs across several models, highlighting how GPT‑6 Luna is among the cheapest ever offered by OpenAI. The author also comments on visual differences in model outputs, noting that GPT‑6 outputs are more muted compared to GPT‑5.6.