Topics
Latest
AI
Amazon
Image Credits:Jaap Arriens/NurPhoto / Getty Images
Apps
Biotech & Health
Climate
Image Credits:Jaap Arriens/NurPhoto / Getty Images
Cloud Computing
DoC
Crypto
Enterprise
EVs
Fintech
Fundraising
Gadgets
gage
Government & Policy
computer hardware
layoff
Media & Entertainment
Meta
Microsoft
Privacy
Robotics
security measure
societal
Space
inauguration
TikTok
fare
Venture
More from TechCrunch
case
Startup Battlefield
StrictlyVC
Podcasts
video
Partner Content
TechCrunch Brand Studio
Crunchboard
Contact Us
OpenAI declare on Wednesday the launching of o3 and o4 - miniskirt , new AI reasoning models designed to pause and work through questions before responding .
The troupe calls o3 its most advanced logical thinking model ever , outperforming the company ’s previous models on tests measuring math , coding , reasoning , science , and visual understanding capableness . Meanwhile , o4 - mini offers what OpenAI says is a competitive craft - off between price , speed , and public presentation — three cistron developers often regard when choosing an AI model to power their applications .
The new models are part of OpenAI ’s effort to beat out Google , Meta , xAI , Anthropic , and DeepSeek in the cutthroat global AI wash . While OpenAI was first to release an AI reasoning model , o1 , competitors apace followed with versions of their own that lucifer or exceed the carrying out of OpenAI ’s lineup . In fact , reasoning models have start to dominate the arena as AI labs look to eke more performance out of their systems .
O3 nearly was n’t let go in ChatGPT . OpenAI CEO Sam Altman betoken in February that the company intended to dedicate more resources to a advanced alternative that integrate o3 ’s engineering science . But private-enterprise pressure seemingly goad OpenAI to reverse course in the end .
OpenAI says that o3 accomplish state - of - the - art execution on SWE - bench verified ( without customs staging ) , a test measure coding power , hit 69.1 % . The o4 - miniskirt simulation achieves similar operation , scoring 68.1 % . OpenAI ’s next adept model , o3 - mini , mark 49.3 % on the test , while Claude 3.7 Sonnet score 62.3 % .
OpenAI claims that o3 and o4 - mini are its first models that can “ think with double . ” In practice , users can upload epitome to ChatGPT , such as whiteboard sketches or diagram from PDFs , and the model will analyse the images during their “ chain - of - cerebration ” phase before answering . Thanks to this newfound power , o3 and o4 - miniskirt can understand blurry and depressed - quality images and can perform tasks such as zooming or rotating image as they conclude .
Join us at TechCrunch Sessions: AI
Exhibit at TechCrunch Sessions: AI
Beyond image - processing capabilities , o3 and o4 - miniskirt can run and put to death Python code direct in your web browser via ChatGPT ’s Canvas feature article , and search the web when asked about current events .
In improver to ChatGPT , all three model — o3 , o4 - miniskirt , and o4 - mini - high — will be available via OpenAI ’s developer - facing endpoints , the Chat Completions API and Responses API , allow railroad engineer to build applications with the company ’s manikin at usage - free-base rate .
OpenAI is charging developers a relatively grim price for o3 , give its improve performance , at $ 10 per million input tokens ( roughly 750,000 word , longer than the Lord of the Rings serial ) and $ 40 per million turnout token . For o4 - mini , OpenAI is charging the same as o3 - mini , $ 1.10 per million input tokens and $ 4.40 per million outturn token .
OpenAI CEO Sam Altman has indicate o3 and o4 - miniskirt may be its last point of view - alone AI reasoning models in ChatGPT before GPT-5 , a model that the ship’s company has say will unify traditional model like GPT-4.1 with its reasoning models .
