github LlamaIndex · 1h ago · agents

Llama Index v0.14.24 adds Claude Sonnet 5, GPT-5.6, and Gemini 3.7 Flash support

from English

Llama Index released version 0.14.24 on August 19, 2026, introducing support for several new large language models and addressing various core functionality bugs.

llama-index-llms-anthropic and llama-index-llms-bedrock-converse add support for Claude Sonnet 5 and include Claude Opus 5 in model allowlists.
llama-index-llms-openai adds GPT-5.6 models to the supported list.
llama-index-llms-google-genai updates the default model to Gemini 3.7 Flash.
Core fixes include preserving oversized leaf nodes in CodeSplitter, respecting mmr_threshold=0 in MMR embedding search, and ensuring all nodes are kept during IngestionPipeline upserts.
The AG-UI protocol integration now supports multimodal user input including images, audio, video, and documents.
Several reader integrations were updated to properly close file handles and HTTP connections.

This update allows users to leverage the latest model capabilities from Anthropic, OpenAI, and Google while improving stability in document ingestion and memory management.

Importance 1/3 Trust 2/3 LlamaIndex Retrieval & RAG

Related articles

media Latent Space · 1d ago · 1 view

Glean uses model routing to cut costs and select optimal LLMs

Driven by rising frontier model costs and the power of open-weight models, Glean employs model routing to dynamically select the most appropriate AI model for each enterprise task. This approach allows the platform to avoid using large language models for simple operations while optimizing performance and expense for complex queries.

lab Hugging Face Blog · 1d ago · 1 view

Sentence Transformers v6 adds MultiVectorEncoder for late interaction retrieval

Sentence Transformers version 6 introduces the MultiVectorEncoder class, enabling ColBERT-style late interaction retrieval directly within the library without requiring external dependencies like PyLate. This addition allows users to load multi-vector checkpoints from various sources, including PyLate and Stanford-NLP formats, using a unified API.

arxiv arXiv cs.CL · 2d ago · 8 views

Ingestion-time defenses fail against coordinated vector retrieval poisoning

A study demonstrates that admission-time filters cannot prevent coordinated poisoning attacks on vector retrieval systems, as geometrically identical patterns exist for both legitimate niche uploads and malicious injections. The researchers show that a small number of unremarkable documents can surround a target query to seize top-k results in BGE-large / BEIR benchmarks.

arxiv arXiv cs.CL · 2d ago · 7 views

Nemotron-3-Embed closes commercial tax gap in MuSiQue retrieval benchmarks

A new audit of multi-hop retrieval benchmarks reveals that NVIDIA's Nemotron-3-Embed-8B has closed the performance gap between commercially licensed embedders and the non-commercial NV-Embed-v2 anchor on the MuSiQue benchmark. Released on July 16, 2026, this model is the only commercially licensed, free-to-self-host option that performs indistinguishably from the academic standard.

media Hugging Face Forums · 2d ago · 4 views

AlphaAvatar v0.6.6 introduces event-driven memory, audio perception, and unified backend contracts

AlphaAvatar v0.6.6 shifts the framework's focus toward clearer runtime contracts and modular capabilities rather than introducing a single large feature. The release expands Environment Memory to include audio segments alongside visual observations and transitions memory updates from periodic polling to an event-driven model.