NeurIPS turns 40 in 2026, and the NeurIPS 2026 program signals a wider brief than a standard paper track. According to the NeurIPS 2026 website, this edition adds or renews calls spanning Evaluations & Datasets, Reproducibility, Position Papers, Creative AI, Educational Resources, Competitions, Tutorials, Workshops, Affinity Events, and even an AI Reviewing Experiment. The conference runs December 6–12, 2026.
What the NeurIPS 2026 program now includes
The official site lists a familiar Call for Papers supported by a Main Track Handbook and reviewing guidelines, but the breadth beyond that is the story here. NeurIPS is treating infrastructure as scholarship. The Evaluations & Datasets track comes with its own FAQ and guidelines, putting measurement and data curation on equal footing with models. The NeurIPS reproducibility track formalizes expectations for code, runs, and claims. There are calls for position pieces, tutorials, competitions, workshops, affinity gatherings, educational resources, creative AI, and social events — each with its own framing and, in several cases, dedicated review guidance. All of these items are enumerated on the NeurIPS 2026 website.
That lineup matters because it shapes what researchers choose to spend a year building. A dataset with thoughtful documentation can change a subfield. A careful benchmark can reveal hidden regressions. Treating these outputs as first-class submissions nudges labs to invest in them.
Why these tracks matter beyond the paper count
The Evaluations & Datasets track can counter a familiar failure mode: progress that appears only on narrow, brittle benchmarks. Independent, well-specified tests help separate real gains from overfitting. Groups like MLCommons have shown how shared measurements can redirect effort; their public benchmark work offers a template for community coordination (MLCommons benchmarks). A conference-sanctioned home for this work expands that impact.
The NeurIPS reproducibility track points the same way. Artifact evaluation and badging standards, long used in computing venues, have made it easier to trust claims and reuse code. The Association for Computing Machinery documents a clear approach to artifact review and badges that many readers will recognize (ACM artifact review and badging). By dedicating review capacity to reproducible research, NeurIPS can reward teams that do the unglamorous work of packaging, verifying, and documenting their results.
The creative AI call underscores that machine learning now touches culture, design, and education. It invites projects that explore how systems shape expression and learning in the wild. That includes failure modes and social context, not just demos.
How AI-assisted review could change decisions
Buried among the calls is a small but consequential line: an AI Reviewing Experiment. NeurIPS is explicit about it on the 2026 site. Automated assistance in peer review is not just about speed. It can change which weaknesses get flagged first, how consistency is enforced, and where attention goes. It might raise the floor on basic checks — missing citations, unclear claims, code that doesn’t run — while risking new biases if models mirror past decisions too closely.
Peer review has long drawn scrutiny for inconsistency and workload pressure. Large venues receive far more submissions than humans can comfortably vet in a short window. AI tools could triage and surface issues that humans then judge, which may improve signal-to-noise if deployed with care. The risk is over-trust. Conferences will need transparency around what assistance is used, where it applies, and how conflicts or model errors are handled. The NeurIPS move puts those questions on the record for 2026, rather than pretending review is unchanged.
There’s a second-order effect here. If review assistance checks claims against code or evaluates statistical clarity, authors will preempt those checks. That can raise the quality bar before submission. Initiatives focused on reproducibility, such as public efforts cataloging code and results across papers (Papers with Code reproducibility), have already nudged behavior in that direction. Formalizing it inside the conference review loop could extend that pressure.
NeurIPS 2026 tracks in context with the conference calendar
Calendar context helps explain the breadth. ICLR 2026, another major machine learning venue, lists calls for papers, workshops, socials, and blog posts on its own site, with the conference scheduled for April 23–27, 2026 (ICLR 2026). That mix highlights how top-tier venues are experimenting with formats that live beyond traditional papers. NeurIPS is now making that expansion explicit across evaluations, datasets, and reproducibility, not just community add-ons.
Affirming tracks for educational resources and affinity events also matters. Teaching materials and community spaces address who gets to participate and how quickly newcomers can contribute. Those outcomes compound over time. By framing them as part of the NeurIPS 2026 program, organizers are telling participants that mentorship, access, and on-ramps count.
What authors and exhibitors should watch next
The NeurIPS 2026 website points to a Main Track Handbook and multiple reviewing guideline documents across tracks. Authors should read the track-specific instructions closely, because criteria will differ for datasets, position papers, or competitions. Exhibitors have a dedicated 2026 portal as well, which suggests the expo component remains a significant part of the week, even as scholarly calls broaden. All of these links and sections are visible on the official NeurIPS 2026 page.
Plan around fixed dates. The conference itself runs December 6–12, 2026. Work backward from that week for camera-ready deadlines, artifact checks, and travel approvals. Teams submitting to the evaluations and datasets track should budget extra time for documentation and licensing review. Groups eyeing the reproducibility track should treat clean packaging and environment capture as first-class tasks, not a sprint at the end.
The big test for the NeurIPS 2026 program is whether these calls produce clearer evidence about what works, and why. If the AI reviewing experiment shortens review cycles without dulling judgment, and if the non-paper tracks pull in rigorous benchmarks and reusable tools, you’ll see it in the papers and demos that land in December.
Forty years in, the conference is quietly changing what counts and how it’s judged. That will shape lab roadmaps, student projects, and even hiring briefs in 2026. On balance, the wider NeurIPS 2026 program feels like a bet that better measurement and clearer review will pay off — in fewer hype cycles, and more results you can build on.