GPT-5 arrived Aug. 7, 20251. It scored 74.9% on SWE-bench Verified and 94.6% on the AIME '25 math test, per OpenAI's own developer post. Two days earlier the company had shipped gpt-oss-120b and gpt-oss-20b2. That was its first open-weight release since GPT-2, under an Apache 2.0 license open enough for firms to tune and host it themselves3. Forty-eight hours, two model families, one research organization proving it could ship a frontier system and an open sibling in the same week. The pace held through 20264. GPT-5.1, GPT-5.2, a coding-focused GPT-5.3-Codex, GPT-5.4 and GPT-5.5 kept arriving every few weeks. One trade outlet clocked six weeks between GPT-5.4 and GPT-5.5 alone. Behind that schedule, a research leadership team was reshaping itself567.
A Launch Built for Two Audiences
GPT-5's pricing shows a company chasing volume as much as status. The flagship runs $1.25 per million input tokens and $10 per million output1. A mini version drops to $0.25 and $2. Nano undercuts both, at five cents in and forty cents out. Context stretches to 272,000 input tokens and 400,000 total, enough for agent work that chains dozens of tool calls in one session. Aider's polyglot coding benchmark put GPT-5 at 88%. OpenAI's own materials leaned on coding and agent tasks as the model's defining strength, rather than chat quality.
gpt-oss answered a different question. Could OpenAI compete with Chinese open-weight labs on their own ground? Meta had given ground as Llama's pace slowed. DeepSeek had shown an open model could rival closed systems on cost. Alibaba's Qwen kept climbing download charts. Shipping gpt-oss-120b and gpt-oss-20b under Apache 2.0, a license open enough to resell, gave OpenAI a credible entry as the stakes rose23. Pairing it with GPT-5's closed launch, two days apart, let the company claim both ends of the market in a week.
Pachocki, Chen and the Cost of Consolidation
Two names sit atop OpenAI's research org. Jakub Pachocki is chief scientist, and Mark Chen is chief research officer6. A July 2025 MIT Technology Review profile put both of them at the wheel of the roadmap that produced GPT-5. Pachocki described his own job in one sentence. "I am responsible for setting the research roadmap and establishing our long-term technical vision." He added a qualifier about how the two divide work. "But there's fluidity in the roles. We're both researchers, we pull on technical threads."
Chen framed what he wants from a model, and refused to call the job finished. "I really don't consider reasoning done. We've definitely not solved it."6 Pachocki named the target more precisely. "This is really what we're after. How do we get models to discover this sort of novel insight?" Read both answers together. One chases novel insight. The other calls reasoning unfinished. They run an organization that ships every few weeks while telling the public the core problem stays open.
Consolidation carried a cost, and Jerry Tworek paid it. Wired's Maxwell Zeff reported on Jan. 9, 2026 that Tworek, the vice president of research, planned to leave after the company sided with Pachocki in a fight over direction5. Read that plainly. A leadership team let a strategy argument end in an exit rather than a compromise. That differs from the collegial tone OpenAI's public materials favor. Tworek had helped build the reasoning line behind o3 and o4-mini earlier in 2025. Leaving six months after GPT-5 shipped points to a fight about where reasoning goes next, apart from whether it had already worked.
Shazeer's Arrival Changes the Roster
Noam Shazeer's move landed five months later, against the direction talent had moved for years. CNBC reported on June 18, 2026 that Shazeer left for OpenAI7. He had co-led Gemini after returning to Google through its 2024 Character.AI licensing deal. Shazeer co-invented the Transformer design that sits under nearly every large language model built since 2017. His exit handed OpenAI a researcher whose work runs straight through the paper the industry cites first.
OpenAI shipped gpt-oss and GPT-5 two days apart in August 2025, then lost Tworek and gained Shazeer by June 2026
Dates of OpenAI's two August 2025 launches and two research-leadership moves, as reported
Each point marks one event on a real time scale: the magenta points are the two launches, two days apart at the left, and the blue points are the research-leadership moves that followed.
Sources: OpenAI [1]; OpenAI [2]; Wired, via Techmeme [5]; CNBC [7]. Chart by Kelly Dennis.
The numbers behind this chart
| Date | Event |
|---|---|
| Aug. 5, 2025 | gpt-oss-120b and gpt-oss-20b ship. Two days before GPT-5, OpenAI's first open-weight release since GPT-2 |
| Aug. 7, 2025 | GPT-5 ships. 74.9% on SWE-bench Verified, per OpenAI |
| Jan. 9, 2026 | Wired reports Tworek plans to leave. The vice president of research, after a fight over direction |
| June 18, 2026 | Shazeer leaves Google's Gemini for OpenAI. CNBC's report; he co-invented the Transformer design |
That arrival lands inside the vacancy Tworek's exit created. A chief scientist and chief research officer consolidate authority. Their vice president of research leaves rather than accept the direction they picked. Then a Transformer co-inventor arrives from the other frontier lab. The sequence reads as OpenAI trading an internal dissenter for an outside name, on a bet that Shazeer's credibility outweighs whatever memory Tworek took with him.
Compute Beneath the Cadence
Every release above runs on infrastructure scaled to match, and OpenAI moved there too. The company and Nvidia announced a partnership to deploy 10 gigawatts of Nvidia systems8. Sam Altman tied that commitment to the whole economy. "Everything starts with compute. Compute infrastructure will be the basis for the economy of the future, and we will utilize what we're building with NVIDIA to both create new AI breakthroughs and empower people and businesses with them at scale."
That figure explains part of why a release pace measured in weeks held. A numbered strategy running GPT-5.1 through GPT-5.5 depends on compute arriving on a matching schedule. Multi-year gaps, like the one between GPT-3 and GPT-4, ask far less of a supply chain. Dedicated coverage of OpenAI's broader silicon and data-center commitments sits elsewhere in this package. The point here is sequencing, rather than scale.
Reading the Pattern
Every piece of this points at one logic. OpenAI treats releases and research leadership as linked, and adjusts both faster than the industry norm. GPT-5 and gpt-oss shipped within a week because the company built two tracks that move on their own. Point releases kept coming through 2026 because compute and headcount grew to support them. Pachocki and Chen made the calls that set the pace. Tworek's exit shows what happens when a senior researcher disagrees. Shazeer's arrival shows OpenAI hiring to replace the exact caliber of researcher a fast schedule risks losing.
Pachocki weighed in on his own roadmap in September 2026. He posted an essay titled "An Alien Mind" to OpenAI's site, and Fortune's Jeremy Kahn covered it the same week9. "No lab has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer," Pachocki wrote. The chief scientist who built the numbered cadence named a ceiling on it himself.
Whether it holds depends on a question the reporting leaves open. Do GPT-5.5 and what follows mark real gains, or smaller changes wearing a familiar name? Pachocki and Chen built a group that ships on time. Holding quality at that speed, with a team that keeps turning over, tests something else.
By the numbers
- A SWE-bench Verified score of 74.9% arrived with GPT-5 at launch1.
- Pricing runs $1.25 and $10 per million input and output tokens, against five cents and forty cents for nano1.
- Context reaches 400,000 tokens in total, pairing 272,000 input with 128,000 reasoning and output1.
- Two days separated gpt-oss's release from GPT-5's launch in August 202512.
- Ten gigawatts: the scale of Nvidia systems OpenAI committed to deploy under its Nvidia partnership8.
- Jan. 9, 2026: Wired's reported timing for vice president of research Jerry Tworek's planned exit5.
- June 18, 2026: CNBC's reported timing for Noam Shazeer's move from Google's Gemini team to OpenAI7.
What to watch
OpenAI's next numbered release tests whether the research group holds the pace Pachocki and Chen set after Tworek left5. Shazeer's first visible work, whenever it shows up in a model card or technical report, will show how fast a researcher of his standing settles into a team run differently from the one he left7. gpt-oss uptake, set against Qwen and DeepSeek download counts, will show whether the open release moved developer attention or matched a gesture other labs made first23.
Sources
- OpenAI, "Introducing GPT-5 for developers," OpenAI, Aug. 7, 2025, https://openai.com/index/introducing-gpt-5-for-developers/
- OpenAI, "Introducing gpt-oss," OpenAI, Aug. 5, 2025, https://openai.com/index/introducing-gpt-oss/
- OpenAI, "gpt-oss-120b & gpt-oss-20b Model Card," arXiv, Aug. 12, 2025, https://arxiv.org/abs/2508.10925
- Wikipedia contributors, "GPT-5.5," Wikipedia, accessed Sept. 4, 2026, https://en.wikipedia.org/wiki/GPT-5.5
- Maxwell Zeff, "OpenAI VP of Research Jerry Tworek is leaving, sources say," Wired, via Techmeme, Jan. 9, 2026, https://www.techmeme.com/260108/p40
- MIT Technology Review Staff, "The two people shaping the future of OpenAI's research," MIT Technology Review, July 31, 2025, https://www.technologyreview.com/2025/07/31/1120885/the-two-people-shaping-the-future-of-openais-research/
- CNBC Staff, "Google Gemini co-lead Noam Shazeer leaves for OpenAI," CNBC, June 18, 2026, https://www.cnbc.com/2026/06/18/google-gemini-co-lead-noam-shazeer-leaves-for-openai.html
- Nvidia, "OpenAI and NVIDIA announce strategic partnership to deploy 10 gigawatts of NVIDIA systems," Nvidia Newsroom, Sept. 22, 2025, https://nvidianews.nvidia.com/news/openai-and-nvidia-announce-strategic-partnership-to-deploy-10gw-of-nvidia-systems
- Jeremy Kahn, "OpenAI lays out its progress towards 'recursive self-improvement'—even as its chief scientist warns of the risks and says he hopes for a slowdown," Fortune, Sept. 8, 2026, https://fortune.com/2026/09/08/openai-rsi-progress-jakub-pachocki-warns-dangers-slowdown-safety-rules/