OpenAI's new embedding models and API updates
text-embedding-3-small was priced far below OpenAI's previous embedding model, and GPT-3.5 Turbo's input price was cut 50% in its third reduction in a year.
- Models & capabilities
- Minor
OpenAI released two new embedding models, text-embedding-3-small and text-embedding-3-large, which convert text into numerical vectors used for search, clustering and retrieval-augmented applications. The smaller model was priced well below OpenAI’s previous embedding offering, while the larger model traded higher cost for improved accuracy on the company’s internal benchmarks — the first update to OpenAI’s embeddings line since its predecessor, released in late 2022.
Alongside the embeddings, OpenAI cut GPT-3.5 Turbo’s API pricing for the third time in a year, halving the input-token price and reducing the output-token price by a quarter, and shipped an updated GPT-3.5 Turbo model. It also released a new GPT-4 Turbo preview model aimed at reducing instances where the model left coding and other tasks incomplete — a complaint that had circulated among developers since GPT-4 Turbo’s November 2023 launch — and a new moderation model, text-moderation-007, described as its most capable yet for flagging harmful text.
The update also introduced new API key management tools, letting developers create multiple keys per project and track usage and billing separately for each, alongside a mechanism to set spending limits at the project level.
None of the individual changes was a capability leap on the scale of a new frontier model, but together they reflected a pattern that continued through the year: incremental price cuts and infrastructure tooling aimed at retaining developers building production applications on OpenAI’s API, in a market where competing embedding and inference providers were undercutting on cost.