openai timeline
on this page
a decade-long arc from non-profit ai safety org to hundred-billion-dollar enterprise. i’ve tried to be comprehensive here, though the selection of events naturally reflects what i think matters.
most of this is documented fact. some of it is informed speculation. caveat lector.
event timeline
event details
alexnet wins imagenet
september 30, 2012: alexnet demolishes the imagenet competition. alex krizhevsky, ilya sutskever, and geoffrey hinton’s cnn achieves 15.3% top-5 error - absolutely crushing the 26.1% second place.
the 10.8 point gap wasn’t incremental progress. it was proof that deep learning worked. relus, gpu training, dropout - the whole modern stack started here. and note the name: ilya sutskever, future openai co-founder, was already pushing the boundaries.
sources:
- ImageNet Classification with Deep Convolutional Neural Networks (NIPS 2012 Paper)
- Official ILSVRC 2012 Results
google acquires deepmind
january 26, 2014: google buys deepmind for somewhere between $400-650 million. the leading independent ai lab was now owned by one of the world’s largest corporations.
for many in the field, this was the moment the agi race went from academic to corporate. the fear wasn’t just about google having demis hassabis and his team - it was about what happens when the path to artificial general intelligence runs through mountain view. this acquisition directly motivated openai’s founding.
sources:
- TechCrunch: Google Acquires Artificial Intelligence Startup DeepMind For More Than $500M
- TIME: Google Buys Secretive A.I. Company DeepMind for $400 Million
facebook opens paris ai lab
june 2015: facebook opens its third fair lab in paris, with yann lecun at the helm. another tech giant staking its claim in the ai talent war.
the partnership with inria was smart - tap into european research talent before google could. by mid-2015, the pattern was clear: every major tech company was building its own ai research empire.
sources:
- CNBC: Why Facebook chose to open its first European A.I. lab in Paris and not London
- Silicon Republic: Facebook chooses Paris for its third AI research lab
google open-sources tensorflow
november 9, 2015: google releases tensorflow to the world. their internal ml system, already powering photos and search, now free under apache 2.0.
this wasn’t charity - it was strategy. control the tools, shape the ecosystem, attract the talent. tensorflow would become the de facto standard for years, cementing google’s position at the center of the ml universe.
sources:
openai founded
december 11, 2015: openai launches as the anti-google deepmind. a non-profit promising to “benefit humanity as a whole, unconstrained by a need to generate financial return.”
sam altman and elon musk as co-chairs, with greg brockman (ex-stripe cto) and ilya sutskever (poached from google) leading technical. they announced $1 billion in commitments - serious money for a non-profit. the message was clear: agi shouldn’t belong to google.
sources:
openai gym launch
april 27, 2016: openai gym goes live. finally, a standard way to benchmark rl algorithms.
before gym, every paper used different environments, making comparisons nearly impossible. now you had atari games, robot simulations, classic control problems - all with the same api. simple idea, massive impact on rl research.
sources:
universe platform release
december 5, 2016: universe drops - turn any program into a gym environment. websites, games, desktop apps, whatever.
the implementation was delightfully hacky: run everything through vnc, let the agent see pixels and control mouse/keyboard. exactly how a human would interact. ambitious vision for general intelligence, though it never quite caught on like gym did.
sources:
openai five project
2017-2019: openai decides to beat humans at dota 2. not chess, not go - a real-time strategy game with partial information, 170,000 possible actions per turn, and games lasting 45 minutes.
the five coordinated agents trained by playing 180 years worth of games against themselves every day. by april 2019, they were ready to take on og, the world champions. spoiler: the bots won.
sources:
- OpenAI Blog: OpenAI Five
- OpenAI Blog: The OpenAI Five Benchmark
- OpenAI Blog: OpenAI Five Defeats Dota 2 World Champions
elon musk resigns from board
february 2018: elon musk leaves the openai board. official reason: avoiding conflicts with tesla’s ai work.
the real story emerged years later: musk had proposed merging openai with tesla, with him in charge. the board said no. in december 2024, openai released emails showing musk wanted majority equity, board control, and to be ceo. when rejected, he suggested openai “attach to tesla as its cash cow” - the only path he saw that could “hold a candle to google.” without dramatic change, he gave openai a 0% chance of success. this was the first crack in the facade - the tension between “democratize ai” and “win the race” was already showing.
sources:
- OpenAI Blog (Feb 2018): Original Departure Announcement (at bottom of post)
- OpenAI Blog (Mar 2024): OpenAI and Elon Musk
- OpenAI Blog (Dec 2024): Elon Musk wanted an OpenAI for-profit
gpt-1 paper published
june 11, 2018: gpt-1 paper drops. “improving language understanding by generative pre-training.”
the recipe was elegant: train on a mountain of text with no labels (unsupervised), then fine-tune on specific tasks. 117m parameters - tiny by today’s standards, revolutionary at the time. this two-stage approach would become the foundation for everything that followed.
sources:
- OpenAI Blog: Improving language understanding with unsupervised learning
- Official Paper: Improving Language Understanding by Generative Pre-Training (PDF)
openai five defeats pro players
august 5, 2018: openai five beats former pros 2-1 in a best-of-three.
later that month at the international (dota’s world championship), they played exhibition matches against active pros. lost those, but the coordination between the five agents was unlike anything people had seen. the bots were learning.
sources:
gpt-2 announced (withheld)
february 14, 2019: gpt-2 announcement - and the first real controversy. 1.5 billion parameters, text so coherent it scared them.
openai’s decision: don’t release the full model. too dangerous for fake news, spam, manipulation. instead, staged release to “study implications.” the open in openai suddenly had an asterisk. the safety vs openness debate had begun in earnest.
sources:
”capped-profit” pivot
march 11, 2019: the pivot. openai creates a “capped-profit” subsidiary - openai lp.
the logic: compute costs were exploding, top talent wanted equity, and google had infinite money. solution: cap investor returns at 100x (lol), everything above goes to humanity. trying to have it both ways - venture scale funding with non-profit vibes. narrator: the cap would not hold.
sources:
openai five defeats world champions
april 13, 2019: openai five defeats og, the actual world champions, 2-0.
first time ai beat the world champion at a major esport. the scale was staggering: 800 petaflop/s-days of compute, 45,000 years of self-play. openai had proven they could throw unprecedented compute at a problem and win. foreshadowing.
sources:
gpt-2 staged release (355m)
may 2019: gpt-2 medium (355m parameters) released to the public.
the staged release continues. let researchers poke at it, find the biases, build detection tools. still holding back the big one though.
sources:
microsoft invests $1 billion
july 22, 2019: microsoft writes a billion-dollar check. openai gets azure credits and cash, microsoft gets exclusive commercialization rights.
the deal: openai runs everything on azure, microsoft gets first dibs on the tech. both sides talking about “agi” and “benefiting humanity.” in hindsight, this was microsoft’s steal of the century.
sources:
- OpenAI Blog: Microsoft invests in and partners with OpenAI
- Microsoft Blog: OpenAI forms exclusive computing partnership with Microsoft
gpt-2 staged release (774m)
august 2019: gpt-2 large (774m) goes public.
openai declares minimal evidence of misuse so far. the full 1.5b model still locked away, but the staged release was working - sort of. the world didn’t end with the 355m model.
sources:
full gpt-2 model released
november 5, 2019: the full gpt-2 (1.5b) finally drops. experiment over.
verdict: “no strong evidence of misuse.” they released everything - weights, code, the works. the safety theater had served its purpose: generating headlines, establishing thought leadership on ai safety, and buying time to figure out the business model.
sources:
gpt-3 paper published
may 28, 2020: “language models are few-shot learners” - gpt-3 enters the chat. 175 billion parameters.
the breakthrough: you didn’t need to fine-tune anymore. just show it a few examples in the prompt and it figured out what you wanted. in-context learning, emergent abilities - the field would never be the same. this was the paper that made everyone realize scale might be all you need.
sources:
openai api launch
june 11, 2020: the openai api launches. gpt-3 as a service.
no more releasing models - now it’s “text in, text out” through their servers. the justification trifecta: need money for research, democratizing access (through…paid api?), and preventing misuse. the transformation from open research to closed product was complete.
sources:
dall-e unveiled
january 5, 2021: dall-e unveiled. 12 billion parameters trained on text-image pairs.
“an armchair in the shape of an avocado” became the meme. but the real story: transformers worked for images too. same architecture, different data, new modality. the path to multimodal models was opening up.
sources:
chatgpt launch
november 30, 2022: chatgpt drops as a “research preview.” just a chat interface on gpt-3.5.
what made it different: free, simple, conversational. no api keys, no documentation, just type and go. openai had accidentally created the iphone moment for ai. the “research preview” would break the internet.
sources:
chatgpt reaches 1 million users
december 5, 2022: sam altman tweets - 1 million users in 5 days.
for context: instagram took 2.5 months. facebook took 10 months. chatgpt did it in less than a week.
sources:
chatgpt reaches 100m monthly users
january 2023: 100 million monthly active users. two months.
fastest growing consumer app in history. tiktok took 9 months to hit 100m. instagram took 2.5 years. the ai boom wasn’t coming - it had arrived.
sources:
microsoft announces $10b investment
january 23, 2023: microsoft doubles down - $10 billion more.
the terms got more interesting: microsoft gets openai tech across azure, bing, office. openai gets the compute to stay ahead. satya nadella had seen the future in those chatgpt growth numbers. google was caught completely flat-footed.
sources:
- Microsoft Blog: Microsoft and OpenAI extend partnership
- OpenAI Blog: OpenAI and Microsoft extend partnership
gpt-4 launch
march 14, 2023: gpt-4 launches. multimodal, passes the bar exam in the top 10%.
but here’s what they didn’t tell you: model size, architecture, training data - all secret now. no more papers with technical details. just benchmarks and api access. $20/month for chatgpt plus or join the api waitlist. openai had gone full product company.
sources:
chatgpt reaches 100m weekly users
november 2023: first devday conference - 100 million weekly active users announced.
people weren’t just trying chatgpt anymore. they were using it every week, integrated into their work. a year after launch, the retention was holding.
sources:
sam altman fired
november 17, 2023: friday afternoon bombshell - the board fires sam altman.
reason: “not consistently candid in his communications.” microsoft found out from twitter. greg brockman stripped of chairman role, quits immediately. the nonprofit board had just blown up a $90 billion company. chaos incoming.
the vague explanation fueled intense speculation, giving rise to the “what did ilya see?” meme. the theory: ilya sutskever, the research purist, had witnessed a dangerous new capability - a glimpse of agi - that convinced him altman’s commercial drive was an existential threat. this narrative was amplified by reports of a secret project, q* (q-star), that had supposedly achieved breakthroughs in mathematical reasoning, a key step toward agi. days before the firing, researchers had reportedly sent a letter to the board warning that a new discovery could “threaten humanity.” while unconfirmed, the story captured the core conflict: the idealists vs. the capitalists.
sources:
- OpenAI Blog: OpenAI announces leadership transition
- Reuters: OpenAI researchers warned board of AI breakthrough ahead of CEO ouster, sources say
- MIT Technology Review: Unpacking the hype around OpenAI’s rumored new Q* model
emmett shear appointed interim ceo
november 19, 2023: emmett shear (ex-twitch ceo) named interim ceo.
the board doubles down. negotiations to bring sam back had failed. they were really going through with this. employees and investors in open revolt.
sources:
employee revolt
november 20, 2023: monday morning madness.
satya announces sam and greg are joining microsoft to lead a new ai lab. then the employee letter drops: 700+ of ~770 employees threaten to quit and follow sam to microsoft unless the board resigns. plot twist: ilya sutskever, who helped orchestrate the coup, signs the letter and tweets his regret. the board had lost.
sources:
- Washington Post: Nearly all OpenAI employees threaten to quit, follow Sam Altman to Microsoft
- Axios: Hundreds of OpenAI staff tell board to resign or they’ll quit
- Ilya Sutskever (X/Twitter): Post expressing regret
sam altman reinstated
november 21-22, 2023: the board surrenders.
“agreement in principle” announced late tuesday. by wednesday, it’s official: sam’s back. new board: bret taylor (chair), larry summers, and adam d’angelo (the lone survivor). the five-day coup was over. silicon valley had its king back.
sources:
microsoft gets board observer seat
november 29, 2023: microsoft secures a non-voting observer seat on openai’s reconstituted board.
the deal gives microsoft visibility into governance without formal voting power. sam altman calls it the “right choice” in a note to staff. the observer seat won’t give microsoft a vote but provides direct insight into board decisions. it’s the price of their $13 billion investment and the insurance policy against future board drama. microsoft had been blindsided by the november coup - never again.
sources:
- CNBC: Microsoft secures nonvoting board seat at OpenAI
- The Washington Post: OpenAI says Microsoft will have non-voting board seat
elon musk sues openai
march 5, 2024: elon files suit against openai, altman, and brockman. claims breach of founding agreement.
the lawsuit alleges openai abandoned its non-profit mission to “benefit humanity” in favor of profits. musk wants the court to force openai to open-source gpt-4 and stop microsoft from profiting. classic elon timing - right as openai’s valuation hits the stratosphere.
sources:
altman reinstated to board
march 8, 2024: wilmerhale investigation concludes - sam joins the board.
findings: “breakdown in trust” but nothing fireable. new directors added: sue desmond-hellmann, fidji simo. the board that fired him was gone, replaced by people who wouldn’t. corporate governance theater complete.
sources:
- TechCrunch: OpenAI announces new board members, reinstates CEO Sam Altman
- CNBC: Sam Altman rejoins OpenAI board, company adds three new members
gpt-4o launch
may 13, 2024: gpt-4o drops. “o” for omni - text, audio, vision, all native.
gpt-4 level performance but faster and cheaper. the kicker: free users get access. after eighteen months of $20/month subscriptions, openai was giving the goods away. growth > revenue, at least for now.
sources:
ilya sutskever departs
may 14, 2024: ilya sutskever, co-founder, chief scientist, and board member, resigns.
six months after his central role in the failed coup, ilya is out. his departure tweet is gracious, praising openai’s trajectory. sam altman calls him “one of the greatest minds of our generation.” the last of the original board’s idealists was gone, marking the final victory of the company’s commercial faction.
sources:
- Ilya Sutskever (X/Twitter): Departure Announcement
- Sam Altman (X/Twitter): Response to Ilya’s Departure
ilya sutskever founds ssi
june 19, 2024: ilya announces safe superintelligence inc. (ssi). a new company with one goal: “safe superintelligence.”
no products, no services, no revenue - just a laser focus on building ssi without the distractions of a large commercial lab. co-founded with daniel gross (ex-apple ai) and daniel levy (ex-openai). the mission is a direct rebuke to the path openai has taken. the prodigal son of ai safety has left the temple to build his own.
sources:
- [SSI (X/Twitter): Company Launch Announcement](https://x.com/ssi/status/18034 SSI 26858 SSI 03456)
- TechCrunch: Ilya Sutskever, OpenAI’s former chief scientist, launches new AI company
microsoft drops board observer seat
july 10, 2024: microsoft relinquishes its observer seat on openai’s board.
eight months after securing visibility into openai’s governance, microsoft backs off. the official line: “significant progress” at openai makes the seat unnecessary. the real story: antitrust regulators in the u.s. and europe were circling. apple, reportedly planning a similar observer role, also backs away. the optics of big tech controlling the ai frontier had become untenable.
sources:
- CNBC: Microsoft drops its observer seat on OpenAI board amid regulatory scrutiny
- Axios: Microsoft gives up observer seat on OpenAI board
musk refiles lawsuit
august 5, 2024: musk withdraws june lawsuit, immediately refiles with federal racketeering claims.
after dropping his state court case in june, elon’s back with rico violations. accuses openai of operating a “racketeering enterprise” to defraud him. the suit now includes microsoft as a defendant. openai calls it “baseless” and notes musk is just mad he’s not in charge.
sources:
mira murati departs
september 25, 2024: mira murati, chief technology officer, resigns.
another high-profile exit. murati, a key figure in the development of chatgpt and dall-e, leaves to “create the time and space to do my own exploration.” her departure signals a changing of the guard and the end of an era for the team that brought generative ai to the mainstream.
sources:
openai releases musk emails
december 2024: openai drops the receipts - years of emails showing musk wanted control.
facing another lawsuit, openai goes nuclear: publishes emails from 2015-2018 showing musk demanded majority equity, board control, and ceo position. when rejected, he suggested merging with tesla as the “cash cow.” best line: without dramatic change, musk gave openai “0% probability” of success. aged like milk.
sources:
deepseek-r1 release jolts market
january 20, 2025: china’s deepseek ai releases deepseek-r1, an open-source model with performance rivaling gpt-4, trained for less than $6 million.
the release sends shockwaves through silicon valley. if china can produce sota models this efficiently, what does that mean for the western hardware monopoly? by january 27, deepseek’s app tops the ios app store, surpassing chatgpt. nvidia’s stock craters 17% in a single day, wiping out $600 billion in market cap. the u.s. chip export controls, designed to slow china’s progress, suddenly look ineffective. america’s ai “sputnik moment” had arrived.
sources:
- DeepSeek API Docs: DeepSeek-R1 Release
- Wikipedia: DeepSeek R1
- CNBC: China’s DeepSeek quietly releases upgraded R1 AI model
musk appointed to doge
january 20, 2025: elon musk is appointed to lead the newly formed department of government efficiency (doge) via executive order.
the appointment solidifies musk’s alliance with the new administration. the department’s name, a nod to his favored cryptocurrency dogecoin, signals his immense influence. tasked with cutting federal spending, musk serves as a “special government employee” with a 130-day limit. the mandate: “large scale structural reform” and $2 trillion in cuts. reality would prove less ambitious.
sources:
- The White House: Establishing And Implementing The President’s “Department Of Government Efficiency”
- Wikipedia: Department of Government Efficiency
stargate initiative announced
january 21, 2025: the white house announces “stargate,” a $500 billion ai infrastructure project, flanked by sam altman, larry ellison, and masayoshi son.
the press event is a show of force, uniting the biggest names in ai and hardware. the plan: build massive data centers across the u.s. to ensure american dominance in ai. larry ellison talks about curing cancer; trump talks about creating 100,000 jobs. first facility already under construction in texas. it’s the military-industrial complex for the ai age, a direct response to china’s growing capabilities and the deepseek wake-up call. musk immediately tweets they “don’t actually have the money.”
sources:
- CNN: Stargate: Trump announces a $500 billion AI infrastructure investment
- Time: What to Know About ‘Stargate,’ OpenAI’s New Venture
- Wikipedia: Stargate LLC
mira murati founds thinking machines lab
february 18, 2025: mira murati launches thinking machines lab, a new ai startup.
five months after leaving openai, murati is back with a powerhouse team, including openai co-founder john schulman. the company raises a massive 12 billion valuation. they promptly reject a multi-billion dollar acquisition offer from meta, signaling their intent to remain independent. the openai diaspora is now a real competitor.
sources:
- The Federal: Who is Mira Murati? Ex-OpenAI CTO turns down Meta’s $1 billion offer
- TechCrunch: OpenAI’s ex-CTO Mira Murati has recruited OpenAI co-founder John Schulman
ex-staff oppose for-profit transition
april 11, 2025: twelve former openai employees file an amicus brief supporting musk’s lawsuit, opposing the company’s for-profit transition.
the brief, filed by harvard law professor lawrence lessig, argues the restructuring “fundamentally violates” openai’s mission. most damning: todor markov (now at anthropic) calls sam altman “a person of low integrity” who “directly lied to employees” about forcing departing employees to sign lifetime non-disparagement agreements. markov admits he stands to lose money if musk wins - “a large fraction of my life’s savings are in openai equity” - but files anyway. the old guard making one last stand for the original mission.
sources:
- TechCrunch: Ex-OpenAI staff file amicus brief opposing the company’s for-profit transition
- Fortune: 12 former OpenAI employees asked to be heard in Elon Musk’s lawsuit
restructures as public benefit corp (pbc)
may 5, 2025: openai transitions its for-profit arm from a “capped-profit” limited liability company into a public benefit corporation (pbc).
the move clarifies the corporate structure while legally embedding the mission. under the new arrangement, the original openai non-profit entity remains in overall control of the for-profit pbc. this structure is intended to allow openai to raise capital more easily while ensuring the company’s primary fiduciary duty is to its mission of benefiting humanity, not just to shareholders.
sources:
musk departs doge
may 28, 2025: after 130 days, elon musk’s tenure at doge expires.
the alliance crumbles. musk’s role as “special government employee” hits its legal limit. his parting shot: doge achieved only 2 trillion. tensions had flared over a “massive spending bill” that musk said “undermines” doge’s work. trump insists musk isn’t really leaving - “he’ll be back and forth” - but the experiment in meme-named governance was effectively over. by july, musk announces the formation of the “america party.”
sources:
- NPR: Elon Musk is leaving the federal government. What’s next for DOGE?
- NBC News: Elon Musk leaves Trump White House after DOGE tenure
- CNN: Musk says he is forming new political party after fallout with Trump
gpt-oss models released
august 5, 2025: plot twist - openai goes open source (again). gpt-oss-120b and gpt-oss-20b under apache 2.0.
why now? meta’s llama was eating their lunch in the open ecosystem. strategy: dominate both sides - best proprietary models for enterprise, competitive open models for developers. the “open” in openai suddenly meant something again, sort of.
sources:
u.s. government partnership (gsa)
august 6, 2025: uncle sam gets chatgpt enterprise. $1 per agency for year one.
the entire federal government on openai’s platform for basically free. not about the money - it’s about embedding so deep in government infrastructure that you become irreplaceable. microsoft playbook, executed perfectly.
sources:
gpt-5 launch
august 7, 2025: gpt-5 arrives. two days after going “open,” they drop the closed crown jewel.
“unified system” with sota everything - coding, reasoning, multimodal, the works. message clear: here’s some nice open models for the community, but if you want the real thing, you pay. classic openai.
sources: