Announced in 2016, Gym is an open-source Python library created to facilitate the advancement of reinforcement learning algorithms. It aimed to standardize how environments are defined in AI research, making released research more quickly reproducible [24] [144] while supplying users with a basic interface for engaging with these environments. In 2022, new advancements of Gym have actually been transferred to the library Gymnasium. [145] [146]
Gym Retro

Released in 2018, Gym Retro is a platform for reinforcement knowing (RL) research study on video games [147] using RL algorithms and research study generalization. Prior RL research study focused mainly on enhancing agents to resolve single tasks. Gym Retro provides the capability to generalize between video games with similar concepts but different appearances.

RoboSumo

Released in 2017, RoboSumo is a virtual world where humanoid metalearning robotic agents at first lack knowledge of how to even walk, however are provided the objectives of finding out to move and to press the opposing representative out of the ring. [148] Through this adversarial learning procedure, the representatives find out how to adapt to altering conditions. When a representative is then gotten rid of from this virtual environment and positioned in a brand-new virtual environment with high winds, the representative braces to remain upright, recommending it had learned how to balance in a generalized method. [148] [149] OpenAI's Igor Mordatch argued that competitors in between representatives could produce an intelligence "arms race" that could increase a representative's ability to function even outside the context of the competition. [148]
OpenAI 5

OpenAI Five is a group of 5 OpenAI-curated bots utilized in the competitive five-on-five computer game Dota 2, that find out to play against human gamers at a high skill level entirely through trial-and-error algorithms. Before becoming a group of 5, the first public presentation took place at The International 2017, the yearly best championship tournament for the video game, where Dendi, an expert Ukrainian player, lost against a bot in a live one-on-one matchup. [150] [151] After the match, CTO Greg Brockman explained that the bot had discovered by playing against itself for two weeks of genuine time, and that the knowing software was an action in the direction of creating software application that can manage complex tasks like a surgeon. [152] [153] The system utilizes a form of reinforcement knowing, as the bots learn gradually by playing against themselves numerous times a day for months, and are rewarded for actions such as killing an enemy and taking map goals. [154] [155] [156]
By June 2018, the capability of the bots expanded to play together as a full group of 5, and they were able to beat teams of amateur and semi-professional gamers. [157] [154] [158] [159] At The International 2018, OpenAI Five played in 2 exhibit matches against professional gamers, but wound up losing both video games. [160] [161] [162] In April 2019, OpenAI Five beat OG, the reigning world champs of the video game at the time, 2:0 in a live exhibition match in San Francisco. [163] [164] The bots' last public look came later that month, where they played in 42,729 total games in a four-day open online competitors, winning 99.4% of those video games. [165]
OpenAI 5's systems in Dota 2's bot gamer shows the difficulties of AI systems in multiplayer online fight arena (MOBA) games and how OpenAI Five has actually demonstrated using deep reinforcement learning (DRL) representatives to attain superhuman skills in Dota 2 matches. [166]
Dactyl

Developed in 2018, Dactyl utilizes device discovering to train a Shadow Hand, a human-like robotic hand, to control physical things. [167] It learns totally in simulation utilizing the exact same RL algorithms and training code as OpenAI Five. OpenAI dealt with the object orientation problem by utilizing domain randomization, a simulation method which exposes the student to a range of experiences rather than attempting to fit to reality. The set-up for Dactyl, aside from having movement tracking electronic cameras, likewise has RGB cams to enable the robotic to control an approximate object by seeing it. In 2018, OpenAI revealed that the system had the ability to control a cube and an octagonal prism. [168]
In 2019, OpenAI demonstrated that Dactyl might solve a Rubik's Cube. The robotic was able to fix the puzzle 60% of the time. Objects like the Rubik's Cube present complicated physics that is harder to model. OpenAI did this by enhancing the toughness of Dactyl to perturbations by using Automatic Domain Randomization (ADR), a simulation method of generating gradually harder environments. ADR varies from manual domain randomization by not needing a human to define randomization ranges. [169]
API

In June 2020, OpenAI revealed a multi-purpose API which it said was "for accessing new AI designs established by OpenAI" to let developers call on it for "any English language AI task". [170] [171]
Text generation

The company has actually promoted generative pretrained transformers (GPT). [172]
OpenAI's original GPT model ("GPT-1")

The initial paper on generative pre-training of a transformer-based language model was composed by Alec Radford and his colleagues, and released in preprint on OpenAI's website on June 11, 2018. [173] It showed how a generative design of language might obtain world knowledge and process long-range dependencies by pre-training on a varied corpus with long stretches of adjoining text.

GPT-2

Generative Pre-trained Transformer 2 ("GPT-2") is an unsupervised transformer language model and the successor to OpenAI's initial GPT model ("GPT-1"). GPT-2 was revealed in February 2019, with only limited demonstrative versions initially released to the general public. The full variation of GPT-2 was not immediately launched due to concern about possible abuse, consisting of applications for writing fake news. [174] Some specialists revealed uncertainty that GPT-2 postured a significant risk.

In reaction to GPT-2, the Allen Institute for Artificial Intelligence responded with a tool to find "neural fake news". [175] Other researchers, such as Jeremy Howard, alerted of "the innovation to absolutely fill Twitter, email, and the web up with reasonable-sounding, context-appropriate prose, which would drown out all other speech and be impossible to filter". [176] In November 2019, OpenAI launched the complete version of the GPT-2 language model. [177] Several websites host interactive presentations of various instances of GPT-2 and other transformer designs. [178] [179] [180]
GPT-2's authors argue unsupervised language designs to be general-purpose learners, shown by GPT-2 attaining cutting edge accuracy and perplexity on 7 of 8 zero-shot jobs (i.e. the model was not more trained on any task-specific input-output examples).

The corpus it was trained on, called WebText, contains somewhat 40 gigabytes of text from URLs shared in Reddit submissions with at least 3 upvotes. It prevents certain problems encoding vocabulary with word tokens by using byte pair encoding. This permits representing any string of characters by encoding both specific characters and multiple-character tokens. [181]
GPT-3

First explained in May 2020, Generative Pre-trained [a] Transformer 3 (GPT-3) is a not being watched transformer language design and the follower to GPT-2. [182] [183] [184] OpenAI specified that the complete version of GPT-3 contained 175 billion specifications, [184] two orders of magnitude bigger than the 1.5 billion [185] in the full version of GPT-2 (although GPT-3 designs with as couple of as 125 million specifications were also trained). [186]
OpenAI stated that GPT-3 was successful at certain "meta-learning" tasks and could generalize the purpose of a single input-output pair. The GPT-3 release paper offered examples of translation and cross-linguistic transfer knowing between English and Romanian, and between English and German. [184]
GPT-3 considerably improved benchmark results over GPT-2. OpenAI warned that such scaling-up of language designs could be approaching or encountering the essential ability constraints of predictive language designs. [187] Pre-training GPT-3 required numerous thousand petaflop/s-days [b] of calculate, compared to tens of petaflop/s-days for the complete GPT-2 model. [184] Like its predecessor, [174] the GPT-3 trained model was not right away released to the public for concerns of possible abuse, although OpenAI prepared to permit gain access to through a paid cloud API after a two-month complimentary private beta that started in June 2020. [170] [189]
On September 23, 2020, GPT-3 was certified exclusively to Microsoft. [190] [191]
Codex

Announced in mid-2021, Codex is a descendant of GPT-3 that has actually in addition been trained on code from 54 million GitHub repositories, [192] [193] and is the AI powering the code autocompletion tool GitHub Copilot. [193] In August 2021, an API was launched in personal beta. [194] According to OpenAI, the design can produce working code in over a lots programs languages, the majority of efficiently in Python. [192]
Several issues with glitches, style flaws and security vulnerabilities were mentioned. [195] [196]
GitHub Copilot has been implicated of giving off copyrighted code, with no author attribution or license. [197]
OpenAI revealed that they would discontinue assistance for Codex API on March 23, 2023. [198]
GPT-4

On March 14, 2023, OpenAI announced the release of Generative Pre-trained Transformer 4 (GPT-4), efficient in accepting text or image inputs. [199] They revealed that the upgraded technology passed a simulated law school bar exam with a rating around the leading 10% of test takers. (By contrast, GPT-3.5 scored around the bottom 10%.) They said that GPT-4 might likewise read, evaluate or create as much as 25,000 words of text, and write code in all significant programming languages. [200]
Observers reported that the iteration of ChatGPT using GPT-4 was an improvement on the previous GPT-3.5-based model, with the caveat that GPT-4 retained some of the problems with earlier modifications. [201] GPT-4 is also efficient in taking images as input on ChatGPT. [202] OpenAI has actually decreased to reveal various technical details and data about GPT-4, such as the accurate size of the design. [203]
GPT-4o

On May 13, 2024, OpenAI announced and released GPT-4o, which can process and produce text, images and audio. [204] GPT-4o attained state-of-the-art lead to voice, multilingual, and vision criteria, setting new records in audio speech recognition and translation. [205] [206] It scored 88.7% on the Massive Multitask Language Understanding (MMLU) benchmark compared to 86.5% by GPT-4. [207]
On July 18, 2024, OpenAI released GPT-4o mini, a smaller variation of GPT-4o changing GPT-3.5 Turbo on the ChatGPT user interface. Its API costs $0.15 per million input tokens and $0.60 per million output tokens, compared to $5 and $15 respectively for GPT-4o. OpenAI anticipates it to be particularly helpful for enterprises, start-ups and developers seeking to automate services with AI agents. [208]
o1

On September 12, 2024, yewiki.org OpenAI released the o1-preview and o1-mini designs, which have actually been created to take more time to think about their reactions, causing greater precision. These models are particularly effective in science, coding, and thinking jobs, and were made available to ChatGPT Plus and Employee. [209] [210] In December 2024, o1-preview was changed by o1. [211]
o3

On December 20, 2024, OpenAI unveiled o3, the follower of the o1 reasoning model. OpenAI also unveiled o3-mini, a lighter and faster variation of OpenAI o3. Since December 21, 2024, this design is not available for public use. According to OpenAI, they are testing o3 and o3-mini. [212] [213] Until January 10, 2025, safety and security researchers had the chance to obtain early access to these models. [214] The design is called o3 rather than o2 to avoid confusion with telecoms services service provider O2. [215]
Deep research study

Deep research is a representative developed by OpenAI, unveiled on February 2, 2025. It leverages the capabilities of OpenAI's o3 design to perform comprehensive web surfing, information analysis, and synthesis, delivering detailed reports within a timeframe of 5 to 30 minutes. [216] With searching and Python tools enabled, it reached an accuracy of 26.6 percent on HLE (Humanity's Last Exam) benchmark. [120]
Image category

CLIP

Revealed in 2021, CLIP (Contrastive Language-Image Pre-training) is a model that is trained to evaluate the semantic similarity in between text and images. It can significantly be utilized for image classification. [217]
Text-to-image

DALL-E

Revealed in 2021, DALL-E is a Transformer design that produces images from textual descriptions. [218] DALL-E uses a 12-billion-parameter version of GPT-3 to analyze natural language inputs (such as "a green leather purse formed like a pentagon" or "an isometric view of a sad capybara") and generate matching images. It can produce images of practical items ("a stained-glass window with an image of a blue strawberry") as well as items that do not exist in reality ("a cube with the texture of a porcupine"). As of March 2021, no API or code is available.

DALL-E 2

In April 2022, OpenAI revealed DALL-E 2, an updated version of the design with more sensible outcomes. [219] In December 2022, OpenAI published on GitHub software application for Point-E, a new rudimentary system for transforming a text description into a 3-dimensional model. [220]
DALL-E 3

In September 2023, OpenAI revealed DALL-E 3, a more powerful model much better able to generate images from complicated descriptions without manual prompt engineering and render intricate details like hands and text. [221] It was launched to the public as a ChatGPT Plus function in October. [222]
Text-to-video

Sora

Sora is a text-to-video model that can create videos based on short detailed triggers [223] along with extend existing videos forwards or backwards in time. [224] It can create videos with resolution as much as 1920x1080 or 1080x1920. The maximal length of produced videos is unknown.

Sora's development group called it after the Japanese word for "sky", to signify its "unlimited imaginative capacity". [223] Sora's technology is an adjustment of the technology behind the DALL · E 3 text-to-image design. [225] OpenAI trained the system utilizing publicly-available videos in addition to copyrighted videos certified for that function, however did not reveal the number or the precise sources of the videos. [223]
OpenAI showed some Sora-created high-definition videos to the general public on February 15, 2024, specifying that it might create videos approximately one minute long. It also shared a technical report highlighting the approaches utilized to train the model, and the design's abilities. [225] It acknowledged some of its drawbacks, including battles mimicing complex physics. [226] Will Douglas Heaven of the MIT Technology Review called the demonstration videos "remarkable", but kept in mind that they should have been cherry-picked and might not represent Sora's normal output. [225]
Despite uncertainty from some scholastic leaders following Sora's public demonstration, significant entertainment-industry figures have actually shown substantial interest in the innovation's capacity. In an interview, actor/filmmaker expressed his awe at the innovation's ability to generate practical video from text descriptions, citing its potential to reinvent storytelling and material production. He said that his enjoyment about Sora's possibilities was so strong that he had actually decided to stop briefly prepare for broadening his Atlanta-based film studio. [227]
Speech-to-text

Whisper

Released in 2022, Whisper is a general-purpose speech recognition model. [228] It is trained on a big dataset of varied audio and is likewise a multi-task model that can carry out multilingual speech acknowledgment along with speech translation and language identification. [229]
Music generation

MuseNet

Released in 2019, MuseNet is a deep neural net trained to forecast subsequent musical notes in MIDI music files. It can generate tunes with 10 instruments in 15 styles. According to The Verge, a song produced by MuseNet tends to begin fairly however then fall into turmoil the longer it plays. [230] [231] In pop culture, initial applications of this tool were used as early as 2020 for the internet mental thriller Ben Drowned to develop music for the titular character. [232] [233]
Jukebox

Released in 2020, Jukebox is an open-sourced algorithm to produce music with vocals. After training on 1.2 million samples, the system accepts a genre, artist, and a snippet of lyrics and outputs song samples. OpenAI stated the songs "reveal regional musical coherence [and] follow conventional chord patterns" but acknowledged that the tunes do not have "familiar larger musical structures such as choruses that duplicate" which "there is a substantial space" in between Jukebox and human-generated music. The Verge stated "It's technologically excellent, even if the outcomes seem like mushy versions of tunes that may feel familiar", while Business Insider specified "surprisingly, a few of the resulting songs are appealing and sound legitimate". [234] [235] [236]
User interfaces

Debate Game

In 2018, OpenAI introduced the Debate Game, which teaches devices to debate toy problems in front of a human judge. The function is to research study whether such a method may assist in auditing AI choices and in developing explainable AI. [237] [238]
Microscope

Released in 2020, Microscope [239] is a collection of visualizations of every significant layer and neuron of 8 neural network designs which are often studied in interpretability. [240] Microscope was created to analyze the features that form inside these neural networks quickly. The models consisted of are AlexNet, VGG-19, different versions of Inception, and various versions of CLIP Resnet. [241]
ChatGPT

Launched in November 2022, ChatGPT is an expert system tool built on top of GPT-3 that provides a conversational interface that permits users to ask questions in natural language. The system then responds with a response within seconds.