IT
Agent frameworks, SDKs and CLIs, open source releases, local inference and coding tools.
Hugging Face unifies generation cache names in Transformers
Support for the old cache argument names ends at different versions for `forward` and `generate`.
Last 7 daysNov 17 – Nov 21
Thinking Machines Lab releases Inkling 2 Mini weights
In Thinking Machines Lab’s tests, a research model scored higher on long-context tasks when its middle layers ran more times.
AI illustrationHugging Face routes language models through OpenRouter
Failed requests are retried with another provider, while support for users' own provider keys ends for language models.1 week agoNov 10 – Nov 15
Hugging Face unifies model export return format
The update brings exported models under a shared interface for sharing and execution, while changing how developers save them.
Hugging Face proposes a shared decision-model interface
The proposed interface brings probabilities for predefined answers and different question types under a common input and output contract.
Motif 4 is due as open weights in first quarter of next year
Predecessor Motif 3 is licensed under `MIT`, but public documents do not state the licence terms for the follow-up model.
2 weeks agoNov 04
IBM's 20B slide model Palette arrives with no stated licence
IBM researchers report second place on the three-domain PresentBench mean, but their blog post does not say where the weights are distributed or what commercial terms apply.
3 weeks agoOct 27 – Oct 31
Nanbeige releases 33B model Nanbeige4.3-33B under Apache-2.0
The Terminal-Bench 2.1 score of 80.9 comes from the developer's own evaluation, and training data sources and minimum VRAM are not confirmed in public documents.
Mistral publishes Large 4 weights under modified MIT licence
Commercial use is allowed, but a model-as-a-service business needs a separate commercial licence.
Liquid AI's LFM3 models run on Linux and Windows with Ollama
The models now run beyond Apple silicon, but the release notes do not give their licence or minimum VRAM.
Ollama adds Liquid AI's LFM3 models, on Apple hardware first
The release notes do not name a licence or give a date for Linux and Windows support.
Oct 2026Oct 15 – Oct 24
Ollama v0.40.11 brings macOS speech generation and d3 models
The release notes state no licence and give no VRAM requirements for each d3 size
Apache-2.0 coding model Flint runs first on macOS via Ollama
The `Apache-2.0` model has 40B parameters, but the release notes give no minimum VRAM or training data sources.
Hugging Face introduces TorchNeuron for AWS's Trainium chips
Hugging Face says running its models on AWS Trainium is easier than ever.
Ollama v0.40.6 introduces q2_0 tensor and file types to GGUF
The release notes do not say what clef is or what VRAM q2_0 requires.
Transformers v5.20.0 breaks code with deprecated cache calls
Code that calls the removed cache methods needs changes, and the release notes do not say how far expert-parallel inference has slowed.
FLUX.3 Image Klein comes to Hugging Face's Diffusers library
Hugging Face announced the addition through the title of a blog post published on 15 Oct 2026.
17 in totalWritten by Jiyul SeoRSS feed