Add image_tokens field to ApiMetaBilledUnits

[yurekami]

Summary

• Added image_tokens field to the ApiMetaBilledUnits model
• Updated merge_meta_field utility to properly merge image_tokens across responses
• Added tests for the new field

Problem

The embedding API returns an image_tokens field in the billed_units response, but the ApiMetaBilledUnits model only defined an images field. This mismatch caused image_tokens to only be accessible through Pydantic's extra attributes, leading to confusion and type-checking issues.

Example from the issue:

print(resp.meta.billed_units.model_dump_json(indent=2))
# {
#   "images": null,
#   "input_tokens": 0.0,
#   "image_tokens": 2716  # <-- This field was not in the model
# }

Solution

Added the image_tokens field to ApiMetaBilledUnits:

image_tokens: typing.Optional[float] = pydantic.Field(default=None)
"""
The number of billed image tokens.
"""

Also updated merge_meta_field in utils.py to properly sum image_tokens when merging embed responses.

Test plan

• Added test_image_tokens_field - verifies field can be set, accessed, and serialized
• Added test_merge_with_image_tokens - verifies image_tokens are properly merged
• All existing tests continue to pass

Note

The file notes it's "auto-generated by Fern from our API Definition". The API definition may also need to be updated to include this field for future regeneration.

Fixes #711

🤖 Generated with Claude Code