# LlamaIndex v0.14.16

- Product: LlamaIndex (https://whatsnew.fyi/product/llamaindex)
- Vendor: LlamaIndex
- Date: 2026-03-10
- Version: v0.14.16
- Original notes: https://github.com/run-llama/llama_index/releases/tag/v0.14.16
- Permalink: https://whatsnew.fyi/product/llamaindex/releases/v0.14.16

What's New is an index, not a publisher: every entry below links to the vendor's own release notes, which are the authoritative source. Entries are labelled where they are hand-curated sample data, pre-releases, or drawn from a secondary source such as a developer blog.

Reuse: the summaries, labels and curation here are © What's New. Quote freely with attribution and a link back; wholesale republication of the corpus is not permitted — terms: https://whatsnew.fyi/terms. The vendors' own release notes remain their publishers'.

---

- **added** — Add token-bucket rate limiter for LLM and embedding API calls
- **added** — Add Multimodal LLMReranker
- **added** — Add optional embed_model to SemanticDoubleMergingSplitterNodeParser
- **added** — Add SlidingWindowRateLimiter for strict per-minute caps
- **added** — Extend vector store metadata filters
- **added** — Add Neo4j user agent
- **added** — Add apoc_sample parameter for large database schema introspection in Neo4j
- **added** — Add User-Agent header for Anthropic API calls
- **added** — Add ModelsLab LLM integration
- **added** — Add gpt-5-chat support in OpenAI
- **added** — Support reasoning_content in OpenAI Chat Completions
- **fixed** — Fix chonkie initialization documentation
- **fixed** — Pass tool_choice through FunctionCallingProgram
- **fixed** — Preserve doc_id in legacy_json_to_doc
- **fixed** — Fix async retry backoff to avoid blocking event loop
- **fixed** — Fix additionalProperties in auto-generated KG schema models
- **fixed** — Respect db_schema when custom async_engine is provided
- **fixed** — Replace blocking run_async_tasks with asyncio.gather
- **fixed** — Preserve docstore_strategy across pipeline runs when no vector store is attached
- **fixed** — Fix FunctionTool not respecting pydantic Field defaults
- **fixed** — Fix MarkdownElementNodeParser to extract code blocks
- **fixed** — Pass system_prompt to ReActChatFormatter template
- **fixed** — Fix partial-failure handling in SubQuestionQueryEngine
- **fixed** — Add bounds check to prevent infinite loop in ChatMemoryBuffer.get()
- **fixed** — Ensure streaming flag reset on exception in CondenseQuestionChatEngine
- **fixed** — Pass through run id correctly
- **fixed** — Raise ValueError when 'model' is passed instead of 'model_name' in BedrockEmbedding
- **fixed** — Respect Retry-After header in OpenAI retry decorator
- **fixed** — Improve handling of reasoningContent in responses from Bedrock Converse requests
- **fixed** — Forward allow_parallel_tool_calls for OpenAI chat completions
- **fixed** — Use constrained decoding for OpenAIResponses structured_predict
- **fixed** — Apply cache_control only to last block to respect Anthropic's 4-block limit
- **fixed** — Properly manage async client lifecycle in Azure Inference to prevent unclosed sessions
- **deprecated** — Deprecate asyncio_module in favour of get_asyncio_module
- **security** — Add RestrictedUnpickler to SimpleObjectNodeMapping to prevent insecure deserialization
- **changed** — OTEL instrumentation enhancements

#### Release Notes

##### [2026-03-10]

###### llama-index-core [0.14.16]

- Add token-bucket rate limiter for LLM and embedding API calls ([#20712](https://github.com/run-llama/llama_index/pull/20712))
- Fix/20706 chonkie init doc ([#20713](https://github.com/run-llama/llama_index/pull/20713))
- fix: pass tool_choice through FunctionCallingProgram ([#20740](https://github.com/run-llama/llama_index/pull/20740))
- feat: Multimodal LLMReranker ([#20743](https://github.com/run-llama/llama_index/pull/20743))
- feat: add optional embed_model to SemanticDoubleMergingSplitterNodeParser ([#20748](https://github.com/run-llama/llama_index/pull/20748))
- fix(core): preserve doc_id in legacy_json_to_doc ([#20750](https://github.com/run-llama/llama_index/pull/20750))
- fix: async retry backoff to avoid blocking event loop ([#20764](https://github.com/run-llama/llama_index/pull/20764))
- Fix additionalProperties in auto-generated KG schema models ([#20768](https://github.com/run-llama/llama_index/pull/20768))
- fix: respect db_schema when custom async_engine is provided ([#20779](https://github.com/run-llama/llama_index/pull/20779))
- fix(core): replace blocking `run_async_tasks` with `asyncio.gather` ([#20795](https://github.com/run-llama/llama_index/pull/20795))
- feat(rate_limiter): add SlidingWindowRateLimiter for strict per-minute caps ([#20799](https://github.com/run-llama/llama_index/pull/20799))
- fix(core): preserve `docstore_strategy` across pipeline runs when no vector store is attached ([#20824](https://github.com/run-llama/llama_index/pull/20824))
- Fix FunctionTool not respecting pydantic Field defaults ([#20839](https://github.com/run-llama/llama_index/pull/20839))
- Fix MarkdownElementNodeParser to extract code blocks ([#20840](https://github.com/run-llama/llama_index/pull/20840))
- security: add RestrictedUnpickler to SimpleObjectNodeMapping (CWE-502) ([#20857](https://github.com/run-llama/llama_index/pull/20857))
- feat: extend vector store metadata filters ([#20861](https://github.com/run-llama/llama_index/pull/20861))
- fix(react): pass system_prompt to ReActChatFormatter template ([#20873](https://github.com/run-llama/llama_index/pull/20873))
- refactor: deprecate asyncio_module in favour of get_asyncio_module ([#20902](https://github.com/run-llama/llama_index/pull/20902))
- fix(core): partial-failure handling in SubQuestionQueryEngine ([#20905](https://github.com/run-llama/llama_index/pull/20905))
- fix: add bounds check to prevent infinite loop in ChatMemoryBuffer.get() ([#20914](https://github.com/run-llama/llama_index/pull/20914))
- fix: ensure streaming flag reset on exception in CondenseQuestionChatEngine ([#20915](https://github.com/run-llama/llama_index/pull/20915))
- fix: pass through run id correctly ([#20928](https://github.com/run-llama/llama_index/pull/20928))

###### llama-index-embeddings-bedrock [0.7.4]

- fix: raise ValueError when 'model' is passed instead of 'model_name' in BedrockEmbedding ([#20836](https://github.com/run-llama/llama_index/pull/20836))

###### llama-index-embeddings-openai [0.5.2]

- Respect Retry-After header in OpenAI retry decorator ([#20813](https://github.com/run-llama/llama_index/pull/20813))

###### llama-index-embeddings-upstage [0.5.1]

- chore(deps): bump the uv group across 47 directories with 3 updates ([#20793](https://github.com/run-llama/llama_index/pull/20793))

###### llama-index-graph-stores-neo4j [0.6.0]

- Add Neo4j user agent ([#20827](https://github.com/run-llama/llama_index/pull/20827))
- feat(neo4j): add apoc_sample parameter for large database schema introspection ([#20859](https://github.com/run-llama/llama_index/pull/20859))

###### llama-index-instrumentation [0.4.3]

- otel instrumentation enhancements ([#20816](https://github.com/run-llama/llama_index/pull/20816))

###### llama-index-llms-anthropic [0.10.11]

- Add User-Agent header for Anthropic API calls ([#20771](https://github.com/run-llama/llama_index/pull/20771))
- fix: apply cache_control only to

_[Truncated at 4000 characters — full notes: https://github.com/run-llama/llama_index/releases/tag/v0.14.16]_