All about training
Fine-tuning, instruction-tuning, RL, prompt tuning, in-context learning.
COLM 2026
The 3rd Conference on Language Modeling — Hilton Union Square, San Francisco, October 6–9, 2026. A tidied, bilingual overview of dates, calls, topics, workshops, people and policies, curated from colm.cc.
3Edition
4Conference days
18Workshops
856Accepted papers
Plan your week
The program, every accepted paper, and all 18 workshops — each on its own page.
About COLM
An academic venue for the study of language modeling, broadly defined.
The leap in general-purpose capabilities from language models is a landmark in the development of AI, and it raises a different set of research questions than related venues typically focus on. COLM is an academic venue focused on the study of language modeling, broadly defined — building a community of researchers across disciplines who understand, improve, and critique the development of LM technology.
Venue & format
- Hilton Union Square, San Francisco, USA
- Main conference: October 6–8, 2026 (three single-track days of invited talks, orals, and posters)
- Workshops: October 9, 2026 (about 18 workshops, 7–9 hours each)
- In-person only — there is no virtual attendance option
How do you pronounce COLM? It is pronounced like "Collum", after the Irish name Colm, which means dove.
Key Dates
Key milestones from the official Dates page on colm.cc. Deadlines are 23:59 AoE unless noted; the full list is on the Schedule page.
- Mar 26, 2026
Abstract submission deadline
Abstracts were due 23:59 AoE on OpenReview; full papers followed five days later.
- Mar 31, 2026
Full paper submission deadline
Final deadline for main-track papers (23:59 AoE).
- Apr 14, 2026
Workshop proposal deadline
Proposals for the Friday workshop day; decisions were announced May 12.
- May 22 – Jun 8, 2026
Rebuttal period
Authors respond to reviews on OpenReview.
- Jun 23, 2026
Workshop contributions (suggested deadline)
Each workshop sets its own dates; see the Workshops page for the exact ones. Accept/reject notifications were mandatory by Jul 24.
- Jul 8, 2026
Paper decision notifications
856 papers accepted, 24 of them as oral presentations.
- Aug 10, 2026
Registration opens
Registration opened at 1:00 PM PDT. In-person only; pricing and cancellation policy are on colm.cc.
- Sep 8, 2026
Sponsor payment & insurance deadline
Final date for sponsors; the sponsor portal stays open until Nov 30.
- Oct 6–8, 2026
Main conference
Three single-track days at Hilton Union Square: keynotes, six oral sessions, six poster sessions and a panel.
- Oct 9, 2026
Workshop day
18 workshops run in parallel, 8:30 AM – 6:00 PM.
- Nov 30, 2026
Careers site closes
The COLM careers site accepts applications from Oct 12 until Nov 30.
Topics of Interest
17 areas — "language model" is meant in the broadest sense. Tap a card for the sub-topics.
All about data
Data for pre-training, post-training, and other training stages.
All about evaluation
Benchmarks, simulation environments, scalable oversight, protocols & metrics, bias/equity/misuse.
All about safety
Security, privacy, misinformation, adversarial attacks & defenses.
Science of LMs
Scaling laws, limitations, emergent capabilities, interpretability, training dynamics.
Compute-efficient LMs
Distillation, compression, quantization, sample- and memory-efficient methods.
Engineering for large LMs
Distributed training & inference, training dynamics, optimization stability.
Learning algorithms for LMs
SFT/RL, unlearning, meta-learning, model mixing, continual learning.
Inference algorithms for LMs
Decoding, reasoning, search, planning.
Mind, brain, philosophy, law & LMs
Cognitive science, neuroscience, linguistics, legal perspectives.
LMs with tools and code
Tool/API integration, LM-driven software engineering.
LMs for everyone
Multilinguality, dialects, low-resource & vernacular languages, value pluralism, fairness.
LMs and the world
Factuality, retrieval-augmented LMs, knowledge models, commonsense, world models.
Multimodal LMs
Perception, audio, time series, and other modalities.
LMs and embodiment
Perception, action, and robotics.
LMs and interactions
Conversation, interactive learning, multi-agent learning.
Diverse domains & novel applications
Medicine, education, science, and beyond.
Workshops
About 18 workshops, all on October 9, 2026 (7–9 hours each). Tap a card for the workshop site and call for papers.
Workshop on Efficient Reasoning
Making LM reasoning cheaper and faster.
Workshop on Agent Behavior
Behavior, evaluation, and analysis of AI agents.
The 2nd Workshop on Lifelong Agent: Learning, Aligning, Evolving
Continual learning, alignment, and evolution of agents.
Non-Autoregressive Language Models for Fast and Flexible Text Generation
Text generation beyond left-to-right decoding.
Actionable Interpretability
Turning interpretability insights into action.
Workshop on Scientific Understanding of Foundation Models
A scientific account of why foundation models work.
Generative AI for the World: Globalizing Tasks, Evaluations, and Systems
Globalizing tasks, evaluations, and systems.
Second Tokenization Workshop
Tokenization and its effect on language models.
AdvML-Frontiers × CoTMA: From Model Security to Compositional Threats in Multi-Agent AI
From model security to multi-agent compositional threats.
LLM/VLM Deployment Opportunities and Risks in Healthcare
Deployment opportunities and risks in healthcare.
Social Sim'26: Fidelity in Applications
Fidelity in social-simulation applications.
Context Beyond the Window: Persistent Knowledge in Language Models
Persistent knowledge in language models.
AI Measurement Science: Toward Rigorous AI Evaluation
Toward rigorous, scientific AI evaluation.
2nd Workshop on Human-Centered Privacy and Security for Language Models
Human-centered privacy and security for LMs.
Learning from Situated and Embodied Interaction
Learning from situated, embodied interaction.
Methods and Opportunities at Small Scale (MOSS)
Methods and opportunities for small-scale research.
Workshop on Responsibly Enabling Data for Foundation Models
Responsible data for foundation models.
LM4Sci 2.0: Language Models for Scientific Discovery
Language models for scientific discovery.
Organization
The organizing committee and the COLM board. Tap a card for the person's homepage.
Yoav Artzi
General Chair · Cornell University / Google DeepMind
Greg Durrett
Senior Program Chair · New York University
Aviral Kumar
Program Chair · Carnegie Mellon University
Yulia Tsvetkov
Program Chair · University of Washington
Yoon Kim
Program Chair · MIT
Robert Hawkins
Workshop Chair · Stanford University
Surbhi Goel
Workshop Chair · University of Pennsylvania
Robin Jia
Community & Accessibility Chair · USC
Anjalie Field
Community & Accessibility Chair · Johns Hopkins University
Angelina Wang
Ethics Chair · Cornell University
Vered Shwartz
Ethics Chair · University of British Columbia
Chris Brown
Organizational Chair · AICONS
Yejin Choi
COLM Board · University of Washington, NVIDIA
Denny Zhou
COLM Board · Google DeepMind
Danqi Chen
COLM Board · Princeton University
Angela Fan
COLM Board · Meta
Alice Oh
COLM Board · KAIST
Yoav Artzi
COLM Board · Cornell Tech, ASAPP
Alexander Rush
COLM Board · Cornell University, Anysphere
Dipanjan Das
COLM Board · Google DeepMind
FAQ
The most common questions about submitting to and attending COLM.
Can I submit to COLM while my paper is under review elsewhere?
No — COLM does not allow double submissions. A full paper under review at COLM cannot be simultaneously under review elsewhere (e.g. ICML/NeurIPS), and any ARR paper at any stage counts as under submission. Double-submission checks run after the full-paper deadline. Concurrent submission to non-archival workshops is fine.
Will there be a short / position / survey paper track?
There is no separate short-paper track and no separate call or reviewing for position papers in 2026, opting for simplicity. There is no special survey track either, but you may submit a survey and it will be routed to appropriate area chairs and reviewers.
Can I edit my submission between the abstract and paper deadlines?
Yes — all fields stay editable until the full-paper deadline, but keep edits minimal to avoid disrupting the program committee's work.
What is the preprint policy?
Non-anonymous preprints (arXiv, social media, websites) are allowed. Reviewers are told not to actively search for them, and encountering one is not a conflict of interest. You may submit work that is already a preprint without citing it, but related prior published work by the authors must be cited with appropriate anonymization.
What is the reciprocal reviewing requirement?
Two requirements. Per-submission: each submission is expected to contribute a reviewer (the form lets you designate one or mark an exemption). Per-reviewer: any author on 4+ submissions is automatically added to the reviewer pool unless already on the program committee or organization. The aim is to scale the reviewer pool with submission volume.
What qualifications are required to review?
Reviewers should have research experience equivalent to a second-year ML grad student and have been a primary author on at least two peer-reviewed papers at related venues (e.g. ACL, EMNLP, ICML, NeurIPS, ICLR). First-time reviewers are strongly encouraged to find a qualified, experienced mentor.
What is the expected reviewing load?
Up to about 5 papers per reviewer.
Posters or oral presentations? And poster size?
All papers are presented as posters; a small number are selected for orals (authors notified around mid-September). Posters are 36 in (H) × 72 in (W).
Can I present remotely or have someone hang my poster?
No remote presentation, and the conference cannot place your poster for you. COLM is in-person only and submission entails a responsibility to present. If you truly cannot attend, the recommended option is to find a colleague who is attending to present on your behalf.
Is there an author registration requirement?
At least one author of each accepted paper must register for the main conference (workshop-only registration does not count). Students need only a student registration, and one registration covers an author even with multiple accepted papers.
Do postdocs / recent grads qualify for the student rate?
If you are a recent graduate publishing under your student affiliation, register as a Student. Other non-industry researchers — including postdocs, faculty, and independent researchers — should register as Academic.
Is there a virtual option? How do I get a visa letter?
COLM is in-person only, with no virtual attendance option. Upon registration you can download an invitation letter for visa purposes.
Is COLM archival?
Yes, COLM is an archival venue. You may also submit to non-archival venues, but not to other archival venues.
What is COLM's policy on the use of LLMs?
COLM 2026 adopts ICLR 2026's LLM policy (modified so minor writing/coding assistance need not be disclosed). Substantive use — originating ideas, writing original content or references, generating data/plots, or evaluation — must be disclosed. In reviewing, ANY LLM use must be disclosed. Authors and reviewers remain responsible for their contributions.
How do you pronounce COLM?
Like "Collum", after the Irish name Colm — which means dove.
Guides & Policies
Official documents on colm.cc. Tap a card to open the authoritative page.
Call for Papers
Scope, 17 topic areas, double-blind review, ethics review, page limits and templates.
Call for Workshops
How to propose a workshop: scope, logistics, proposal format, and reviewing criteria.
Code of Conduct
Conduct expected of all participants; required acknowledgment before participating.
Code of Ethics
Ethical standards binding on the whole submission and reviewing process.
Review Guidelines
Guidance reviewers and area chairs are asked to follow.
AC Guidelines
Expectations and responsibilities for area chairs.
Submission Instructions
Step-by-step instructions for preparing and submitting your paper.
Author Guide
Guidance authors are expected to follow throughout the process.
COI Policy
How conflicts of interest are defined and handled via OpenReview profiles.
LLM Usage Policy
Adopts ICLR 2026's policy: substantive LLM use in research must be disclosed; minor copy-editing or coding assistance need not be. Any LLM use in reviewing must be disclosed.
LaTeX Template (2026)
COLM uses a distinct template. Follow its formatting; check that imported packages don't alter its parameters.
Attend COLM 2026
Registration opened Aug 10, 2026. COLM is in-person only — Hilton Union Square, San Francisco, Oct 6–9. Pricing, cancellation policy and hotel booking are on colm.cc.
Registration & pricing