Roadmap
flinq ships a deliberately small surface today. The items below are committed directions, stated plainly so you can plan against them. Dates are not promised during the beta.
Async batch endpoint
Section titled “Async batch endpoint”An asynchronous POST /v1/batches endpoint, OpenAI-shaped: submit a file of
requests, poll for completion, download the results. It runs offline rather than
in your request path and is billed at a 50% discount on the standard per-token
EUR rate, making it the cost-efficient path for large, non-interactive ingests.
See Bulk & batch for the synchronous recipe you can use today.
The GPU-served production model — shipped
Section titled “The GPU-served production model — shipped”Production serving now runs on dedicated GPU hardware (August 2026). The move brought higher throughput and lower latency — typical single-request latency dropped from seconds to well under a second, and per-request input caps rose (see Models) — without changing the API or the vector space your index is built on.
Image embeddings and cross-encoder rerank — shipped
Section titled “Image embeddings and cross-encoder rerank — shipped”flinq-embed-2.0 (October 2026) embeds images into the same vector space as
text, so a text query finds the figure, drawing or diagram it describes, and
flinq-rerank-2.0 reranks with a cross-encoder whose calibrated
relevance_score can decide that no candidate matches. See Models,
Embeddings and
Rerank.
BauSatz across European languages
Section titled “BauSatz across European languages”BauSatz starts with German, built on real GAEB tender data. The direction is an open, neutral benchmark for construction language across European languages, so models can be compared on the same terms beyond a single market. German is flinq’s deepest-trained language; the 2.0 models were also trained and evaluated on 12 further European languages, and the benchmark is meant to measure that coverage on open terms.
Layout-aware PDF ingestion and OCR
Section titled “Layout-aware PDF ingestion and OCR”The Documents category reads GAEB today. PDF joins it as one coherent release: layout-aware extraction that understands headings and tables (so PDF chunks can be hierarchy-based, under the same versioned strategy contract as GAEB), with opt-in OCR for scanned documents. Deliberately not shipped piecemeal — naive text-layer chunking without layout understanding would be no better than what any client library does locally.
Beyond images: drawings, BIM and IFC
Section titled “Beyond images: drawings, BIM and IFC”The built environment is not only text. Image inputs are the first step; whole drawing sets, BIM geometry and IFC property sets carry meaning that neither text nor a single raster image captures. Reading them extends flinq from a domain embedder toward the full foundation model for the built environment.