Topics
later
AI
Amazon
Image Credits:Bryce Durbin / TechCrunch
Apps
Biotech & Health
clime
Chart comparing small AI models from Artificial Analysis. Price here is a combination of input and output tokens.Image Credits:Artificial Analysis
Cloud Computing
Commerce Department
Crypto
endeavour
EVs
Fintech
fundraise
widget
Gaming
Government & Policy
Hardware
layoff
Media & Entertainment
Meta
Microsoft
Privacy
Robotics
Security
Social
distance
inauguration
TikTok
Transportation
Venture
More from TechCrunch
upshot
Startup Battlefield
StrictlyVC
newssheet
Podcasts
Videos
Partner Content
TechCrunch Brand Studio
Crunchboard
reach Us
OpenAI introduced GPT-4o miniskirt on Thursday , its latest small AI mannequin . The company saysGPT-4o miniskirt , which is chinchy and faster than OpenAI ’s current cut - edge AI models , is being released for developers , as well as through the ChatGPT web and mobile app for consumers , starting today . Enterprise users will arrive at access next workweek .
The company says GPT-4o mini outperforms manufacture - lead small AI models on logical thinking tasks require text edition and vision . As small AI models amend , they are becoming more pop for developer due to their speed and toll efficiencies liken to larger mannikin , such asGPT-4 OmniorClaude 3.5 Sonnet . They ’re a useful selection for gamy book , bare tasks that developers might repeatedly call on an AI model to perform .
GPT-4o mini will supervene upon GPT-3.5 Turbo as the small modelling OpenAI offers . The company claims its new AI framework scores 82 % on MMLU , a bench mark to value reasoning , compared to 79 % for Gemini 1.5 Flash and 75 % for Claude 3 Haiku , accord to data fromArtificial Analysis . On MGSM , which measures math logical thinking , GPT-4o mini scored 87 % , compared to 78 % for Flash and 72 % for Haiku .
Further , OpenAI sound out GPT-4o mini is importantly more affordable to run than its previous frontier models , and more than 60 % cheaper than GPT-3.5 Turbo . Today , GPT-4o miniskirt supports text and vision in the API , and OpenAI says the model will corroborate video and audio capabilities in the future .
“ For every corner of the world to be indue by AI , we need to make the models much more low-cost , ” said OpenAI ’s head of intersection API , Olivier Godement , in an interview with TechCrunch . “ I think GPT-4o mini is a really grownup step forward in that direction . ”
For developer construct on OpenAI ’s API , GPT4o mini is price at 15 cent per million input signal token and 60 penny per million output keepsake . The model has a setting window of 128,000 token , roughly the length of a book , and a cognition cutoff of October 2023 .
OpenAI would not disclose exactly how large GPT-4o mini is , but said it ’s roughly in the same tier as other small AI models , such as Llama 3 8b , Claude Haiku and Gemini 1.5 Flash . However , the company arrogate GPT-4o mini to be firm , more price - effective and wise than industry - conduct small models , base pre - launch testing in the LMSYS.org chatbot scene of action . former independent examination seem to sustain this .
Join us at TechCrunch Sessions: AI
Exhibit at TechCrunch Sessions: AI
“ Relative to corresponding models , GPT-4o mini is very degenerate , with a medial output speed of 202 tokens per minute , ” said George Cameron , Co - Founder at Artificial Analysis , in an e-mail to TechCrunch . “ This is more than 2X faster than GPT-4o and GPT-3.5 Turbo and represents a compelling offer for speed - dependent use - cases include many consumer applications and agentic approaches to using LLMs . ”
OpenAI’s new tools for ChatGPT Enterprise
Separately , OpenAI announced new tool for enterprise customers on Thursday . In ablog post , OpenAI announce the Enterprise Compliance API to serve business concern in highly regulated industries such as finance , healthcare , legal services and government comply with logging and audit requirements .
The company enounce these tool will allow admins to audit and take action at law on their ChatGPT Enterprise data . The API will provide record of time - boss interaction , including conversations , uploaded file , workspace users and more .
OpenAI is also giving admins more gritty ascendence for workspace GPTs , a customs duty version of ChatGPT created for specific business purpose cases . Previously , admins could only fully allow or bar GPT activity created in their workspace , but now , workspace owners can create an approved list of domains that GPTs can interact with .
