Notes on the new 1m context window for Claude Sonnet 4: https://
simonwillison.net/2025/Aug/12/cl
aude-sonnet-4-1m/
… You need to send a beta header of context-1m-2025-08-07 and be on tier 4, which means you have purchased at least $400 in API credits
LLMS
-

Claude Sonnet 4 1M Context Window Access Requirements
By
–
-

LFM2-VL Models: 1.6B and 450M Vision-Language Released
By
–
We also provide an inference and a fine-tuning Colab notebooks. LFM2-VL-1.6B: https://
huggingface.co/LiquidAI/LFM2-
VL-1.6B
… LFM2-VL-450M: https://
huggingface.co/LiquidAI/LFM2-
VL-450M
… -

Liquid Releases Fast 450M and 1.6B Parameter VLM Models
By
–
Liquid just released two 450M and 1.6B param VLMs! They're super fast and leverage SigLIP2 NaFlex encoders to handle native resolutions without distortion. Available today on @huggingface
! -

Mistral Medium 3.1 Released with Performance Improvements
By
–
Introducing Mistral Medium 3.1. Overall performance boost, tone improvement, smarter web searches. Try it now in Le Chat (default model) or via our API (`mistral-medium-2508`).
-
Open-source world models welcome announcement
By
–
Welcome to open-source world models! https://t.co/IXVxZdGm4r
— clem 🤗 (@ClementDelangue) 12 août 2025Welcome to open-source world models!
-
CulturalGround: Extracting More Training Data from Cultural Knowledge Bases
By
–
Importantly, CulturalGround is further evidence that we are not out of data yet and we can squeeze out more and more training data from the knowledge bases. By just rethinking where cultural data resides and systematically generating factual questions and answers about the
-

CulturalGround Fine-tuning Achieves State-of-the-Art Performance on Cultural Benchmarks
By
–
To demonstrate the effectiveness of CulturalGround, we fine-tune an existing multimodel model Pangea on a subset of the dataset. The resulting model achieves state-of-the-art performance for its model size on multiple cultural benchmarks in PangeaBench and other benchmarks. Below
-

CulturalPangea Excels in Multilingual and English Performance
By
–
CulturalPangea not only achieves good performance on multilingual cultural understanding, but also perform well on English, and overall.
-

CulturalGround: Building Inclusive Multimodal LLMs for Global Contexts
By
–
Current multimodal LLMs excel in English and Western contexts but struggle with cultural knowledge from underrepresented regions and languages. How can we build truly globally inclusive vision-language models? We are introducing CulturalGround, a large-scale dataset with 22M
-

LLM Performance Drops on Non-Standard Medical Questions
By
–
Another example of a persistent problem with LLMs. They do very well on standard medical questions, but when the right answer is replaced with “none of the above” performance drops. More recent models generally have lower drops in performance. https://
jamanetwork.com/journals/jaman
etworkopen/fullarticle/2837372
…