NeurIPS 2026: Dates, Venue, and What to Expect in Sydney

NeurIPS 2026 runs December 6-12 at a Sydney venue neurips.cc has not yet named, with satellite conferences in Atlanta and Paris on December 9-13. Registration opens July 20 and pricing is unpublished (checked 2026-08-24, neurips.cc). Attend for the papers that will land in production LLMs in 12 to 24 months.

I track AI events because the ML research that ships at venues like NeurIPS becomes the SaaS tools I test 6 to 18 months later. Retrieval-augmented generation, mixture-of-experts, DPO. All of them showed up at NeurIPS before any founder was pitching them. That is the argument for going. Not the trade show. The paper pipeline.

Two things you will read about NeurIPS 2026 that are wrong. First: it is in the United States. It is not. NeurIPS moved the main site from San Diego to Sydney for 2026, with Atlanta and Paris kept on as parallel satellites. Second: a specific registration price is posted. It is not. NeurIPS’s own 2026 pricing page still reads “Pricing is not yet available” as of 2026-08-24, and the portal shows the registration window opens July 20.

Sydney is not a satellite this year. neurips.cc lists Sydney as “Main” for 2026 for the first time. Atlanta stays on as a satellite; Paris replaces Mexico City. If your NeurIPS mental model is “the San Diego conference,” update the map before you book anything.

Sydney is confirmed, the specific venue is not

I keep seeing ICC Sydney at Darling Harbour cited as the venue. It is Sydney’s only harbourside convention center at NeurIPS’s scale, so it is a well-informed guess. But neurips.cc names only the city, not the building. Any source claiming a Sydney venue name with certainty this early is filling in a blank neurips.cc has not filled in itself. I am flagging ICC Sydney as likely, not locked. Book the dates. Wait on the flights.

Who NeurIPS 2026 is worth the trip for

You will get real value in Sydney if your day job is training foundation models, running eval methodology work, or building agent frameworks that ship into product in 2027. NeurIPS accepted papers are a reliable early signal for what lands in production AI systems 12 to 24 months later. Faster R-CNN, this year’s Test of Time Award winner, is a decade old and still underlying most modern object detection. That is what a NeurIPS paper looks like when it compounds.

You will not get value if you are a growth marketer, a product manager without an ML background, or an investor looking for the next round of pitches. Content is peer-reviewed, dense, and built around poster sessions and spotlight talks that assume a machine learning foundation. NeurIPS is a research venue wearing a conference badge, not a marketing event.

The specific trap for 2026: if you budgeted for San Diego based on last year, you now need Sydney airfare, a hotel in a market that runs hot in early December, and a work visa if you are a non-Australian speaker on the program. Block December 6-12. Set a July 20 reminder for registration. Do not book flights until neurips.cc names the venue and posts pricing.

What NeurIPS has actually published about the 2026 program

I checked neurips.cc against every third-party guide currently ranking on Google. NeurIPS does not run a vendor keynote, so there is no headline speaker to wait on. What is confirmed is the organizing structure. General Chairs Hsuan-Tien Lin (National Taiwan University) and Razvan Pascanu (Google DeepMind and Mila) lead the 2026 edition. Finale Doshi-Velez (Harvard) is Senior Program Chair, with Jenna Wiens, David Rolnick, Nika Haghtalab, and Marc Deisenroth as Program Chairs (source: neurips.cc/Conferences/2026, checked 2026-08-24). Everything else, including invited talks, the workshop list, and the session catalog, has not been published yet.

One dated program note is worth flagging. NeurIPS is running a voluntary AI-Assisted Reviewing Experiment this cycle, built into OpenReview. Consenting reviewers get randomly assigned to no LLM assistance, open-ended LLM assistance, or structured LLM assistance per paper, with area chairs blind to which condition produced which review. It is IRB-approved through the University of Texas at Austin and framed as a study, not a policy change. Outside the study, LLM use in reviewing still gets you a desk rejection.

The Pangram episode: 178 desk rejections for likely AI text

I have been watching how peer review responds to LLM-drafted submissions since 2024. The Pangram episode is the loudest public answer so far. In June 2026, the Position Paper Track chairs published an audit run through an AI-detection company called Pangram. 178 submissions (18.4% of the track) were desk-rejected without appeal. Another 123 (12.7%) were told to submit a “pre-AI, post-AI, final” version history proving human authorship by June 15, 2026, or face the same fate.

Submissions flagged at a 90%+ Pangram score jumped from 11.9% in 2025 to 42.7% in 2026 in the Position Paper Track, and from under 1% to 9.3% in the separate Evaluations and Datasets track. That is the atom of NeurIPS 2026. Not a projection, a published number: 178 papers desk-rejected for suspected LLM authorship in a single track, with no right of appeal (source: NeurIPS Position Paper Track chairs, June 2026). It is the first time a top-tier AI research venue has publicly acted at this scale on the exact problem the research it accepts is now creating.

The 2026 author timeline

Abstracts closed May 4, 2026. Full papers May 6. Reviews went out to authors July 22. The author, reviewer, and area-chair discussion window runs July 27 to August 3, followed by a reviewer-and-AC-only window August 3 to 10. Final author notifications land September 24, 2026 (source: neurips.cc/Conferences/2026/Dates). If you submitted this year, September 24 is the only date that decides anything. Everything between now and then is discussion, not verdict.

Registration pricing has not been published

NeurIPS’s 2026 pricing page reads “Pricing is not yet available for this conference” as of 2026-08-24. The registration portal is closed, with an opening date of July 20 and no dollar figures attached. For historical context only: a widely cited 2023 breakdown put fees at roughly $375 for students, $800 for academics, and $1,000 for industry attendees. That is three years old and pre-dates two rounds of NeurIPS attendance growth. Treat it as a floor, not a forecast.

Physical registration has historically included the virtual pass at no extra cost, with a cheaper virtual-only pass also available. If NeurIPS’s usual cancellation policy holds, expect a full refund window into November, dropping to zero after, with a carve-out for visa-denied attendees to convert to virtual and refund the difference. None of that is confirmed for 2026 yet.

Would I book this year?

I would go if I were a working AI or ML researcher, especially one shipping into foundation model, retrieval, agent, or eval methodology work. I would skip if I were a growth marketer, a product manager without an ML background, or an investor looking for pitches. That is not a hedge. NeurIPS is the most-cited AI research venue on the planet and the strongest early indicator of what production ML looks like 18 months out. It is also, by design, hard to follow if you are not already inside the research conversation.

One insight worth carrying: NeurIPS’s Main Track submissions grew from roughly 13,400 in 2024 to 21,575 in 2025 (source: NeurIPS Program Committee post-mortem, September 2025). That is roughly a 61% jump in a single year. It is not from more researchers. It is from LLMs making paper drafting cheaper. The Pangram episode is what happens when that curve meets peer review. Whichever side of that line you sit on, NeurIPS 2026 is where the field publicly decides how to grade it.

(Sydney in early December runs around 25°C with harbour views, so if you have to be trapped inside a conference hall, worse fates exist.)

Frequently Asked Questions

When and where is NeurIPS 2026?

NeurIPS 2026 runs December 6-12 with the main conference in Sydney, plus satellites in Atlanta and Paris running December 9-13. neurips.cc has confirmed the city but not the specific Sydney venue building.

Is NeurIPS 2026 in Sydney and not the United States?

Yes. NeurIPS’s 2026 conference page lists “Main: Sydney, Australia” for the December 6-12 dates. Sydney has hosted as a satellite in previous years but is promoted to main site for 2026. Atlanta stays on as a satellite; Paris replaces Mexico City.

How much does NeurIPS 2026 registration cost?

Not published. neurips.cc’s 2026 pricing page reads “Pricing is not yet available” as of 2026-08-24, and registration opens July 20. A 2023 reference point put fees near $375 (student), $800 (academic), and $1,000 (industry), but that number is three years old.

Who is speaking at NeurIPS 2026?

The organizing committee is confirmed, led by General Chairs Hsuan-Tien Lin and Razvan Pascanu. Individual invited-talk speakers and the workshop list have not been announced.

What happened with the AI-generated papers at NeurIPS 2026?

In June 2026, the Position Paper Track chairs published an audit using Pangram, an AI-detection tool. 178 submissions (18.4%) were desk-rejected for likely AI-generated text with no right of appeal. Another 123 (12.7%) had to submit documented proof of human authorship by June 15, 2026.

How does NeurIPS 2026 compare with the other big ML conferences?

NeurIPS, ICML, and ICLR are the “big three” ML research venues. I covered ICML 2026 in Seoul earlier this year. NeurIPS has the largest submission volume of the three (21,575 in 2025), and its Test of Time Awards routinely surface papers now underlying the best AI tools shipping today.