# What changed in llama-index

> 84 changes in llama-index 0.14 from its official release notes, upgrade guides and advisories: 7 breaking, 1 security, 15 new APIs and options. Each one cited. Sources checked 2026-10-19. HTML: https://unstale.sh/pypi/llama-index/

To see which of these apply to a repository, run `npx @unstale/cli init` in it.

## llama-index 0.14

### Breaking changes

- Frontend tool results are no longer persisted server-side. (0.14.25) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.25)
- The deprecated ipex-llm and optimum-intel IPEX integrations were removed. (0.14.25) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.25)
- The Python version floor was raised to keep vulnerable dependency forks alive. (0.14.25) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.25)
- Raises ValueError instead of fabricating a tool_call_id. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Migrates the MCP tools integration to MCP 2.x. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- The Exa tool was updated to current API conventions, so developers may need to align their usage with those conventions. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)
- The llama-index-memory-mem0 integration upgrades to mem0 2.x. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)

### Security

- Stops hardcoding verify=False for S3 TLS. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)

### New APIs and options

- StructuredLLMRerank now supports native async operations. (0.14.25) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.25)
- Supports user input containing images, audio, video, and documents. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Adds Claude Opus 5 to the Anthropic and Bedrock Converse model allowlists. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Adds Claude Sonnet 5 support to the Anthropic integration, including function calling. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Adds GPT-5.6 models to the supported OpenAI models. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Adds async support. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Accepts any AsyncDBChatStore. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Adds raise_on_error to LLM path extractors. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Supports the thinking type 'disabled' in Bedrock Converse. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Expands support for the V2 API. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Adds Claude Fable 5 support for Anthropic and Bedrock Converse. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)
- Adds Claude Opus 4.8 support for Anthropic and Bedrock Converse. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)
- Adds multimodal query engines. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)
- Adds multimodal synthesis capabilities to core. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)
- Supports the vLLM reasoning field in OpenAI messages. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)

### Behavior and default changes

- The Perplexity integration adds an attribution header. (0.14.25) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.25)
- The Weaviate client is tagged with a LlamaIndex integration telemetry header. (0.14.25) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.25)
- Uses Gemini 3.7 Flash by default in the library. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Tool output parsing now handles DocumentBlock and VideoBlock. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)
- The embeddings and LLM integrations allow google-genai 2.0+ SDKs. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)
- The Gemini Live voice agent integration now allows google-genai SDK versions 2.0 and newer. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)
- Serializes rich content blocks in tool results. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)

### Fixes and other changes

- Compact and refine streaming was restored. (0.14.25) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.25)
- The node_ids query filter now uses Filter.by_id(). (0.14.25) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.25)
- Failed function tools are no longer retried. (0.14.25) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.25)
- Images are now kept on Gemini function responses. (0.14.25) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.25)
- Gemini tool request IDs were fixed. (0.14.25) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.25)
- Metadata replacement now falls back when the target value is None. (0.14.25) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.25)
- Images and files are now kept on OpenAI Responses API tool results. (0.14.25) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.25)
- Persists frontend tool messages. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Isolates copies of the initial state. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Honors an agent's structured_output_fn and output_cls. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Makes aioboto3 optional for the Bedrock embeddings integration. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Makes aioboto3 optional for the Bedrock Converse integration. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Preserves falsy metadata values when indexing nodes. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Validates that the embedding payload is non-empty before making an AWS call. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Preserves multiblock chat history writes. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Gives each citation node its own ID and offsets. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Preserves oversized leaf nodes instead of dropping them. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Sets chunk_size to 32000 and fixes the input format. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Reports the model's effective context window in the llama-cpp integration. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Corrects schema, vector ranking, filters, and batching. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Stops sending default parameters. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Returns None instead of a list when an error is swallowed. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Closes evaluator file handles. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Skips empty nodes during parsing. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Keeps all nodes per document when performing upserts. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Repairs an image ingestion crash and an embedding_model typo. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Fixes the default endpoint and metadata. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Defaults the condition to AND when it is None. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Respects mmr_threshold=0 rather than treating it as unset. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Corrects the Opus 4.6 context window to 1M. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Supports Pinecone client v8 and v9. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Restores fallback to supplied LLM metadata. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Preserves falsy shard_identifier values, such as 0. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Restores compatibility with qdrant-client 1.19.0. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Sends query_history in the DashScope retriever payload. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Uses word tokenization for stopword removal. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Persists non-ASCII characters without escaping them. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Populates response text when writing a streaming chat response to memory. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Avoids overwriting thought signatures. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Does not mark *args or **kwargs as required tool parameters. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Returns real collection properties as node metadata. (0.14.24) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.24)
- Preserves the IndexNode object during model dumping. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)
- Uses the running event loop in the ingestion pipeline. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)
- Deep-copies workflow initial state to prevent mutation leaks across runs. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)
- Resolved an issue involving string_expr and filter precedence in the Milvus vector store. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)
- Fixes matching of missing metadata for NE and NIN filters. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)
- Guards against ZeroDivisionError when input sequences are empty. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)
- Fixes PromptMode import compatibility in the RankLLM reranker integration. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)
- Fixes reuse of kwargs when refreshing reference documents. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)
- Fixes RecursionError when units are larger than chunk_size. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)
- Streaming tool_kwargs are parsed from a string into a dictionary. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)
- Preserves source nodes returned by TreeSelectLeafRetriever. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)
- Preserves URL-backed video and document memory blocks. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)
- The Weaviate vector store now supports newer batch context managers. (0.14.23) — [release notes](https://github.com/run-llama/llama_index/releases/tag/v0.14.23)