Chroma official releases
Chroma software version 1.5.10.dev314 released on October 8, 2026.
The latest version of Chroma is 1.5.10.dev314, released on October 8, 2026.
The complete AI news desk
The AI world, in one verified news room.
Models, agents, robotics, research, business and policy, with the original source attached. Search and move beyond the latest thirty updates.
Open the 30-day AI Change IndexLatest and archive
Open the fast flashes streamInfrastructure
Hugging Face Hub Python SDK official releases
Hugging Face CLI `hf jobs stats` now prints a snapshot by default.
`hf jobs stats` now prints a snapshot by default, exiting immediately.
Weaviate official releases
Weaviate 1.40.0 adds user isolation with namespaces.
Namespaces add user isolation on shared clusters. HFresh optimizes memory usage and disk writes.
Prefect official releases
Prefect 3.8.8 drops Python 3.10 support.
Prefect 3.8.8 will be the last release supporting Python 3.10. It fixes a crash caused by invalid Retry-After headers.
Weaviate official releases
Parquet export row groups now limited to 32 MiB in v1.38.19
The v1.38.19 release limits Parquet row groups to 32 MiB to manage writer memory. It also allows exporting Parquet blob columns without page header statistics.
vLLM official releases
DeepSeek-V4.1-Flash now defaults to NVFP4 compressed KV cache on SM100.
v0.31.0 includes 717 commits from 307 contributors, 96 of whom are new. It features a new vllm preload CLI for fast restarts.
ONNX official releases
Version 1.23.2 fixes Opset 28 bug.
Version 1.23.2 fixes a version miss for Opset 28.
Weaviate official releases
LSM store performance improved by segment replay in v1.39.9
Version 1.39.9 improves LSM store performance with large write-ahead log replay. It also allows backups without collections.
LiteLLM official releases
LiteLLM adds breached password detection to password policy.
LiteLLM v1.104.0 includes breached password detection and self-service password change features.
Dagster official releases
Dagster 1.13.25 adds mobile layout for UI, changes default partition selector.
The UI now includes a mobile layout for phones, with new navigation menus and mobile versions of key pages.
Weaviate official releases
Weaviate 1.39.8: Multi-tenant collections now pass tenant to re-vectorize check on PATCH
The update includes a fix for handling corrupt commit log segments and a new endpoint for near-vector search.
Weaviate official releases
Weaviate 1.38.18: Multi-tenant collections now pass tenant to re-vectorize check on PATCH.
The update includes a fix for the MMR query vector normalization and a patch for multi-tenant collections re-vectorization.
OpenVINO official releases
OpenVINO toolkit 2026.4.1 adds causal-attention support for large language models.
OpenVINO 2026.4.1 fixes GPU issues with remote output tensors and GRU models. It also addresses NPU crashes on specific Intel processors.
Hugging Face Hub Python SDK official releases
Hugging Face Jobs gain retry, rescheduling, and port exposure features.
Hugging Face Jobs now support automatic retries and port exposure.
Prefect official releases
Prefect releases 3.8.8.dev2 with updated dependencies and bug fixes.
The latest nightly development release includes updated dependencies and improved cron schedule help text.
Weaviate official releases
MUVERA Adds 4-bit Rotational Quantization in v1.40.0-rc.2
The release candidate for v1.40.0 introduces general availability of namespaces and vector index dropping.
Microsoft DeepSpeed official releases
Muon's update size remains unchanged under fp16 loss scaling.
Muon's update size remains consistent across loss scales with this change.
ONNX official releases
ONNX Build Bug Fixes Included in Release 1.23.1
Release 1.23.1 fixes a bug in the ONNX build's save_external_data function.
Intel Neural Compressor official releases
Intel Neural Compressor 3.10 adds selective quantization for Keras/JAX.
Intel Neural Compressor 3.10 adds experimental PyTorch support for NVFP4 with E5M3 scaling.
Weaviate official releases
RAFT boot-time race fixed in v1.39.7
RAFT boot-time race fix includes atomic raft node publishing. Batch delete cap exposed in gRPC BatchDeleteReply.
GOV.UK cyber security news
UK Proposes AI Safety, Transparency Measures at UNSC
The Foreign Secretary Ed Miliband spoke at the UNSC on AI's potential risks and benefits. He emphasized the need for government involvement in AI safety and transparency.
NVIDIA Model Optimizer official releases
ModelOpt 0.47.0 introduces FP8 Vision Encoder recipes for qwen3_vl and qwen3_5 models.
ModelOpt 0.47.0 introduces quantization with Autotune for ONNX models, benchmarking placements in requested runtime precision.
Safetensors official releases
The 3.14t Python wheels are now built and uploaded in the release workflow.
The new release includes a prefetch loader for faster big model load times, reducing them from over an hour to under a minute.
FlashAttention official releases
PyTorch 2.13+ now supports C++20 in CUDA extension builds.
New beta release includes CUDA extension build with C++20 for PyTorch 2.13+.