dmodel.ai · Jul 2026
dmodel says LLM agents discovered six concept-erasure algorithm families across 50 concepts and outperformed LEACE on nonlinear probes, reinforcing its interpretability and alignment focus.
Company research
Public company, workplace, funding, and market signals
Updated Jul 29, 2026
Overview
dmodel is a San Francisco-based AI research lab focused on interpretability and alignment, building RL environments and agent workflows that help frontier labs inspect, steer, and evaluate model internals.
Primary product
RL environments and agent workflows for model interpretability/alignment research
Founded
2024
Headquarters
San Francisco, California, United States
Team size
10-20
Work style
Onsite
Industry
Research Services
Sub-industry
AI research, interpretability, and alignment
Offices
San Francisco, CA 94107 (HQ/SOMA office)
San Francisco, California, United States
Business model
research services
custom research collaborations
B2B services
Funding
Stage
pre-seed
Total raised
$500K
Latest round
Pre-Seed · Sep 2024
Latest amount
$500K
Sep 2024 · Y Combinator
People & investors
Leadership
CEO, Co-founder
COO, Co-founder
Adam Scherlis
Founding Research Scientist
Curry Winter
Chief of Staff
Work & culture
Research-first and high-autonomy, with a strong emphasis on frontier-lab partnerships, interpretability work, and in-person collaboration at the SOMA office.
Work style
Onsite
Compensation
Public job postings cite $120k-$170k base pay for an Operations Manager role, plus bonus, benefits, and equity. No broader public compensation bands were found.
Benefits
bonus
benefits
equity
Product & market
Pricing
Custom / bespoke research engagements; no public pricing disclosed.
Differentiators
- —Frontier-lab partnerships
- —Open-ended RL environments for interpretability tasks
- —Agent-based research workflows
- —Independent internal research alongside client work
- —Reproducible technical demos and research writeups
Technology
OpenAI
PyTorch
Customers
frontier labs
AI research teams
Competitors
Redwood Research
Conjecture
Fiddler AI
Truera
Scale AI
CTGT
ModelOp
Scale & contact
Estimated revenue
about $330K ARR (reported 2025)
Estimated monthly visits
5.1K
Traffic estimate as of Aug 2025
Latest news
dmodel.ai · Apr 2025
An accessible follow-up explaining the nullability work, probes, and reading-diagram demo for understanding internal model representations of code semantics.
dmodel.ai · Mar 2025
A research post on probing how models represent program-value nullability, including a 15-program microbenchmark and evidence that models develop an internal concept of nullability as they scale.
dmodel.ai · Aug 2024
An early research post showing how steering vectors can change character behavior more subtly than prompting or fine-tuning, with runnable examples and notebook-based demos.
Jobs and careers at dmodel
0 jobs at dmodel
No open positions
Check back later for new openings