OpenAI continues to rapidly iterate and improve its industry-leading AI models, announcing several major updates today that deliver more value at lower costs for developers.
The company unveiled two new text embedding models that boast stronger performance and significantly reduced pricing compared to previous versions. Additionally, OpenAI shared plans to upgrade its popular GPT-3.5 Turbo next week, with the new version providing upgraded accuracy and multilingual support, as well as slashing input token prices by 50% and output token prices by 25%.
New Embedding Models Set New Standard for Price and Performance
The new text-embedding-3-small and text-embedding-3-large models aim to make state-of-the-art embeddings accessible to more developers by dramatically improving the price-performance ratio.
Compared to the text-embedding-ada-002 model (which was last updated in December 2022), the smaller text-embedding-3-small reduces costs by 5x to $0.00002 per 1k tokens while still boosting accuracy on key benchmarks by over 10 percentage points.
Meanwhile, the aptly named text-embedding-3-large represents OpenAI's most capable embedding model to date. Benchmark scores climb over 20 percentage points higher than its predecessor, cementing text-embedding-3-large as the new gold standard, priced reasonably at $0.00013 per 1k tokens.
Both models introduce native support for shortening generated embeddings to user-specified dimensions, enabling fine-grained balancing of accuracy versus compute and storage costs.
GPT-4 Is No Longer Lazy
Over the last few months, many developers and ChatGPT users began complaining that GPT had suddenly become "lazy". OpenAI says they have addressed this with a new gpt-4-0125-preview model that "completes tasks more thoroughly than the previous preview model "
The company also shared that they plan to launch GPT-4 Turbo with vision in general availability in the coming months.
GPT-3.5 Turbo Update Is Even Cheaper
An updated GPT-3.5 Turbo model ( gpt-3.5-turbo-0125) set to launch next week, will also see 1 significant price reduction, marking the third price cut in the last 24 months. Input costs will be reduced by 50% to $0.0005 per 1k tokens and output prices will drop 25% to $0.0015 per 1k tokens.
OpenAI says the new GPT-3.5 Turbo model will have "various improvements including higher accuracy at responding in requested formats and a fix for a bug which caused a text encoding issue for non-English language function calls."
API Key Management and Content Safety
OpenAI rounded out its slate of announcements with several other impactful updates across critical areas like content safety and API management:
- A new text moderation model (
text-moderation-007) is available for free via OpenAI's Moderation API and is the company's most robust moderation model to-date. - New tools for developers to manage API keys and understand API usage. These tools allow for more granular control and tracking of API usage, facilitating easier management of API features across teams and projects
With relentless improvement across accuracy, multilingual support, safety, and affordability paired with scalable, managed infrastructure, OpenAI continues to present a compelling value proposition for businesses to leverage its models as a service rather than self-hosting their own.