multimodal

What three kinds of data can a message's content contain in LangChain?

A plain string; a list of provider-format blocks (the native format of a specific model); or a list of LangChain standard blocks (content_blocks).

What happens to a message's `content` when you specify `content_blocks` at initialization?

`content_blocks` is just a convenient wrapper — the actual data still ends up populating the message's `content`.

Why does content format flexibility matter for multimodality?

Modern models accept not just text but images; audio; and video; and a list of blocks lets you pass all of that in one message.

Why does LangChain support both native provider formats and its own standard format for content?

Because different APIs (OpenAI; Anthropic; Google) use different formats for multimodal data; so LangChain lets you write in either the provider's native format or its own standard one and converts as needed.

What problem does `content_blocks` solve regarding model reasoning across providers?

Different providers return different formats for the same concept — e.g. Anthropic calls it `thinking` (with a `signature`) while OpenAI calls it `reasoning` (or `reasoning_content`) — so without standardization you'd need separate code per provider.

How does the `content_blocks` property behave when accessed on a message?

It lazily parses `content`; converting the provider format into the standard LangChain format only when it's accessed.

When an Anthropic `thinking` block is converted via `content_blocks`; what does its type become?

`reasoning`.

When an Anthropic `thinking` block is converted via `content_blocks`; where does the `signature` field end up?

Inside `extras`; e.g. `extras: {\signature\': \'...\'}`.'

What three benefits does `content_blocks` provide; according to the notes?

Unified code (write handling once; works across OpenAI; Anthropic; Google; etc.); type safety (returns strictly typed objects; not raw dicts); and laziness (conversion only happens on access; avoiding wasted resources).

By default; are `content_blocks` persisted into `content` during serialization?

No — by default they are not persisted into `content` during serialization; such as when saving to JSON or a database.

What must you do if an external application needs the standard `content_blocks` format saved directly into `content`?

Enable `v1` mode; via the `LC_OUTPUT_VERSION` environment variable or the `output_version` parameter at model initialization.

After enabling `v1` mode; what will a message's `content` directly contain?

Standard blocks instead of the provider format.

python from langchain.messages import HumanMessage # 1. Just a string msg = HumanMessage('Hello!') # 2. Native OpenAI format (provider-specific) msg = HumanMessage(content=[ {'type': 'text'; 'text': 'What's in this image?'}; {'type': 'image_url'; 'image_url': {'url': 'https://example.com/image.jpg'}} ]) # 3. Standard LangChain format (type-safe) msg = HumanMessage(________=[ {'type': 'text'; 'text': 'What's in this image?'}; {'type': 'image'; 'url': 'https://example.com/image.jpg'}; ])

```python from langchain.messages import HumanMessage # 1. Just a string msg = HumanMessage('Hello!') # 2. Native OpenAI format (provider-specific) msg = HumanMessage(content=[ {'type': 'text'; 'text': 'What's in this image?'}; {'type': 'image_url'; 'image_url': {'url': 'https://example.com/image.jpg'}} ]) # 3. Standard LangChain format (type-safe) msg = HumanMessage(content_blocks=[ {'type': 'text'; 'text': 'What's in this image?'}; {'type': 'image'; 'url': 'https://example.com/image.jpg'}; ]) ```

python from langchain.messages import AIMessage # Message from Anthropic in NATIVE format message = AIMessage( content=[ {'type': 'thinking'; 'thinking': '...'; 'signature': 'WaUjzkyp...'}; {'type': 'text'; 'text': '...'}; ]; response_metadata={'model_provider': 'anthropic'} ) # Access content_blocks — get the STANDARD format message.________

```python from langchain.messages import AIMessage # Message from Anthropic in NATIVE format message = AIMessage( content=[ {'type': 'thinking'; 'thinking': '...'; 'signature': 'WaUjzkyp...'}; {'type': 'text'; 'text': '...'}; ]; response_metadata={'model_provider': 'anthropic'} ) # Access content_blocks — get the STANDARD format message.content_blocks ```

python # Option 1: via environment variable import os os.environ['________'] = 'v1' # Option 2: at model initialization from langchain.chat_models import init_chat_model model = init_chat_model('gpt-5-nano'; ________='v1')

```python # Option 1: via environment variable import os os.environ['LC_OUTPUT_VERSION'] = 'v1' # Option 2: at model initialization from langchain.chat_models import init_chat_model model = init_chat_model('gpt-5-nano'; output_version='v1') ```

python message.content_blocks # [ # {'type': '________'; 'reasoning': '...'; 'extras': {'signature': 'WaUjzkyp...'}}; # {'type': 'text'; 'text': '...'} # ]

```python message.content_blocks # [ # {'type': 'reasoning'; 'reasoning': '...'; 'extras': {'signature': 'WaUjzkyp...'}}; # {'type': 'text'; 'text': '...'} # ] ```

python msg = HumanMessage(content=[ {'type': 'text'; 'text': 'What's in this image?'}; {'type': '________'; 'image_url': {'url': 'https://example.com/image.jpg'}} ])

```python msg = HumanMessage(content=[ {'type': 'text'; 'text': 'What's in this image?'}; {'type': 'image_url'; 'image_url': {'url': 'https://example.com/image.jpg'}} ]) ```