Gemini 2.5 Flash
gemini-2.5-flash
Gemini 2.5 Flash is a language model. Called on Mixpeek through the provider API.
Mixpeek calls the provider's API by name. Provider terms apply.
New opens the Designer in Studio. For an existing namespace, open it in Studio and use its Designer tab.
No demo for Gemini 2.5 Flash yet
A video, image, slideshow or playground goes here once we record one on our own data.
What it is good and bad at
- Best at
- The extractors' default for structured extraction and vision work: video description and OCR in multimodal_extractor v2, web page extraction, PDF layout (document_graph), scrolling-text OCR.
- Worst at
- It is a thinking model: the thinking budget can eat the output tokens on long extractions (llm_extractor.py comments); slower and costlier than gemini-3.5-flash-lite.
- Task
- Generation
- Input
- Text, Image, Video, Document
- Output
- Text, Structured data
- Use cases
- Video understanding, Tagging, Captioning, OCR, Brand detection
Templates that use it
- Footage Intelligence at Archive ScaleDeploy this template
Millions of scenes you can move through instead of search. Footage, ads and editor assets land in three clean collections, and scenes are clustered one partition at a time so every run stays under the sample-size caps.
Input and output
Input
- text
- prompt plus content
- image
- page images, frames, panoramas
- video
- segments up to 120 s
Output
- text
- free text, or JSON against response_shape / response_schema
How Mixpeek wires it
- Extractors
multimodal_extractor@v2fixed (description and ocr_text when run_video_description or run_ocr)web_scraper@v1llm_model unset (default), also forced when a blocked expensive model is nameddocument_graph_extractor@v1vlm_model unset (default gemini-2.5-flash)text_extractor@v1response_shape set and llm_model unset (llm_provider google)scrolling_text_extractor@v1fixed (panorama OCR via CentralLLMService)
- Feature URIs
- None today
- Vector indexes
- None
- Retriever stages
- None
- Combines with
- multimodal_extractor v2: Gemini description and OCR text -> E5 description and OCR vectors, and the description feeds the Gemini Embedding 2 vector
How it runs
- Status
- Provider API Provider API, called from extractor batch steps through CentralLLMService or the Gemini video client; universal__api routes it for POST /v1/inference.
- Serve apps
universal__api- Runtime
- Gemini developer API or Vertex AI (GoogleModel.GEMINI_2_5_FLASH, GoogleProvider)
- Weights
- Provider API, no remote code
- Parameters
- Not recorded
- Size on disk
- Not recorded
- Preprocessing
- Prompt per extractor; structured output forced to JSON when OCR and description must be split.
- Postprocessing
- JSON parsed into columns (description, ocr_text, extracted_data).
- Serving
- batch
- Hardware
- Provider API
Scale to zero
- Minimum replicas
- 0
- Maximum replicas
- 6
- Idle timeout
- 600 s
- Memory
- Not recorded
- CPUs
- 0.1
- GPUs
- 0
- Cold start, node warm
- Not recorded
- Cold start, new node
- Not recorded
Measured cost
Billed per token by Google; no committed throughput and cost measurement.
Our evaluations
We have not evaluated this model on our own data yet. Publisher results are in the sources below.
Questions
- What is Gemini 2.5 Flash best at?
- The extractors' default for structured extraction and vision work: video description and OCR in multimodal_extractor v2, web page extraction, PDF layout (document_graph), scrolling-text OCR.
- What are the limits of Gemini 2.5 Flash?
- It is a thinking model: the thinking budget can eat the output tokens on long extractions (llm_extractor.py comments); slower and costlier than gemini-3.5-flash-lite.
- Can Gemini 2.5 Flash be used commercially?
- Yes. Paid Google API under Google's terms; tenants can bring their own key.
- How does Gemini 2.5 Flash run on Mixpeek?
- Provider API, called from extractor batch steps through CentralLLMService or the Gemini video client; universal__api routes it for POST /v1/inference.