COLM 2026 — Conference on Language Modeling

COLM 2026, the third Conference on Language Modeling, runs 6-9 October 2026 at the Hilton San Francisco Union Square (333 O'Farrell Street, San Francisco, California). COLM is an academic venue created specifically for language-model research, and it takes the term 'language model' in the broadest sense: its call for papers spans seventeen subject areas including training approaches, data for each training stage, evaluation methods and benchmarks, safety and adversarial robustness, scaling laws and interpretability, computational efficiency, engineering for large-scale systems, inference and reasoning algorithms, multimodality, embodied AI and robotics, interactive and multi-agent systems, and domain applications in medicine, education and science. The format is deliberately unusual for a machine-learning conference of this size: three days of single-track programming mixing invited talks, oral paper presentations and poster sessions, followed by a fourth day given over entirely to workshops on 9 October. Announced keynote speakers for 6-8 October include Chelsea Finn, Jennifer Pan and John Langford. Papers were submitted through OpenReview with a rebuttal period, and childcare is provided on site by KiddieCorp.

Date

October 6–9, 2026

Location

Hilton San Francisco Union Square, 333 O'Farrell Street, California

Format

In-Person

Event Type

Conference

Status

Upcoming

Pricing

Paid

Organizer

Conference on Language Modeling (COLM)

Added

Jun 24, 2026

Updated

Aug 13, 2026

• ✓Machine learning and NLP researchers
• ✓Applied research scientists at AI labs
• ✓ML engineers working on model training and evaluation
• ✓PhD students and postdocs in language modeling
• ✓Head of AI Research / research managers
• ✓Model evaluation and AI safety practitioners
• ✓Corporate research and university recruiting leads

For a leader running an applied AI or research function, COLM is the highest signal-to-noise technical venue on language models specifically — no exhibition hall, no vendor keynotes, one track. It is where evaluation methodology, safety results and efficiency techniques are argued in public a year or more before they appear in a product, which is exactly the lead time an internal platform roadmap needs. The single-track format and the four-day length make it unusually good for recruiting: your research engineers will see the same talks as the authors they want to hire, and the workshops day is where the newest sub-fields get their first serious hearing.

• ✓Three days of single-track programming: invited talks, oral paper presentations and poster sessions
• ✓Dedicated workshops day on 9 October
• ✓Keynote speakers announced for 6-8 October including Chelsea Finn, Jennifer Pan and John Langford
• ✓Seventeen call-for-papers subject areas spanning training, data, evaluation, safety and interpretability
• ✓OpenReview submission with a formal rebuttal period
• ✓Subject areas covering embodied AI, robotics and multi-agent systems
• ✓Subject areas covering interdisciplinary work in cognitive science, neuroscience, philosophy and law
• ✓On-site childcare provided by KiddieCorp

• ★Single-track for three days — every attendee sees the same talks, which is rare at this scale and makes the hallway conversation converge instead of fragmenting across parallel rooms
• ★Purpose-built for language modeling rather than a general ML conference with an LM track, so training, data, evaluation, safety and interpretability sit on one programme
• ★Explicitly interdisciplinary call for papers: cognitive science, neuroscience, philosophy and law are named subject areas alongside systems engineering and scaling laws
• ★A full fourth day dedicated to workshops (9 October), with its own proposal and submission timeline
• ★On-site childcare provided by KiddieCorp, which materially changes who can attend a four-day conference

COLM has not published its pass prices on its public site. Registration opened to the general public at 1pm Pacific on Monday 10 August 2026 and is handled at colm.eventhosts.cc; check that registration page for current tiers, and expect the usual academic split between student, academic and industry rates.

When is COLM 2026?

COLM 2026 takes place October 6–9, 2026 in San Francisco, California.

Where is COLM 2026 held?

COLM 2026 is held in San Francisco, California. The event format is in-person.

How much does COLM 2026 cost?

COLM 2026 is paid. COLM has not published its pass prices on its public site. Registration opened to the general public at 1pm Pacific on Monday 10 August 2026 and is handled at colm.eventhosts.cc; check that registration page for current tiers, and expect the usual academic split between student, academic and industry rates.

Who should attend COLM 2026?

COLM 2026 is aimed at Machine learning and NLP researchers, Applied research scientists at AI labs, ML engineers working on model training and evaluation, PhD students and postdocs in language modeling, Head of AI Research / research managers, Model evaluation and AI safety practitioners, Corporate research and university recruiting leads.

What topics does COLM 2026 cover?

COLM 2026 covers Language model training and fine-tuning, Evaluation, benchmarks and metrics, AI safety, security and adversarial robustness, Interpretability and scaling laws, Inference, reasoning and planning algorithms, Computational efficiency, distillation and quantization, Multimodal and embodied models, Interactive and multi-agent systems, Multilinguality, low-resource languages and fairness.

This page was written from 5 sources, 2 on domains other than colmweb.org.

1. 1.colmweb.org — colmweb.orgvendor
2. 2.colmweb.org — datesvendor
3. 3.colmweb.org — cfpvendor
4. 4.colm.eventhosts.cc — colm.eventhosts.cc
5. 5.conferencedeadlines.com — colm