The 4 Subtypes of Autism (2025 Research) and What They Mean for Your Child

Autism researchers have known for decades that what we call "autism" is probably several different things bundled under one diagnostic label. The behavioral pattern is consistent enough to be recognizable, but the underlying biology, developmental trajectories, and support needs vary so much that a single category often doesn't capture what's actually going on with any specific child. Researchers call this heterogeneity, and it's the central puzzle in autism science.

The 2025 research from Princeton and the Simons Foundation is one of the more rigorous attempts to actually disentangle this heterogeneity using large datasets and machine learning. Their work identifies four distinct subtypes of autism based on combined behavioral, developmental, and genetic data drawn from the SPARK research database, which has thousands of autistic individuals and their families.

This post walks through what the research found, what it means in practical terms, and what it doesn't mean for your child today.

If you're newer to the autism diagnosis and trying to understand the broader picture, our signs of autism in toddlers pillar covers what autism looks like, and our autism levels explained post covers the level designation system that's used in clinical practice.

---

Why the Subtypes Question Matters

A diagnosis of autism currently tells you that your child shares behavioral patterns with other autistic people: differences in social communication, restricted or repetitive behaviors, sensory profiles. It does not tell you which specific developmental trajectory your child is on, what biological mechanisms produced the autism, or which interventions are most likely to help.

This isn't because clinicians or researchers are bad at their jobs. It's because autism really does encompass a huge range of presentations. A nonspeaking child with intellectual disability and a verbally precocious child with anxiety and rigid thinking are both correctly labeled autistic. They share enough common features for the diagnosis to apply, but the practical reality of supporting them is quite different.

If autism turns out to be three or four or six biologically distinct conditions that look superficially similar, then:

• Different subtypes may respond best to different interventions
• Prognosis information could become more accurate
• Genetic counseling could become more useful for families
• Drug development could target subtype-specific mechanisms (current pharmacology rarely treats core autism features because they vary too much across people)

The subtypes question, in other words, is the question of whether the next generation of autism research can be sharper than the current one. The 2025 work is one piece of that puzzle.

---

What the 2025 Research Actually Did

The Princeton-Simons team used machine learning to analyze data from the SPARK research cohort, the largest single dataset of autistic individuals in the United States. The data included:

• Detailed behavioral profiles (social communication, repetitive behaviors, sensory differences, language, cognitive ability)
• Developmental history (when milestones were hit, when diagnosis occurred, regressions)
• Genetic data (whole-exome or microarray)
• Co-occurring conditions (anxiety, ADHD, intellectual disability, epilepsy, sleep issues)

The machine learning approach (specifically, a clustering algorithm that groups similar profiles together without being told what to look for) identified four distinct groups in the data. Each group has its own combination of behavioral characteristics, developmental trajectory, and genetic signature.

This matters because:

1. The groups weren't predefined by researchers. They emerged from the data, which is more rigorous than a clinical team deciding on subtypes by intuition.
2. The groups are stable across statistical replications.
3. The genetic signatures are different between groups, which suggests the groups may reflect real biological differences rather than just statistical artifacts.

The work has limits, which the authors discuss directly: SPARK is a US research cohort with all the demographic and selection biases that implies, the genetic data is incomplete for most participants, and four groups is the result of one clustering approach (other approaches might find three or six). The findings need replication in other datasets and clinical confirmation before they're considered settled.

---

The Four Subtypes

Different papers describe these groups slightly differently, and the labels are still being refined. The general pattern that's emerged:

Group 1: Social and Behavioral Challenges (largest group)

Roughly 35 to 40% of the studied population. Characterized by:

• Social communication and restricted/repetitive behavior challenges as the dominant features
• Relatively typical or above-average cognitive development
• Later age at diagnosis on average (often after age 5)
• Higher rates of co-occurring anxiety, ADHD, and depression
• Less frequent intellectual disability
• More likely to be missed in early screening, particularly in girls

This group includes many late-diagnosed children and some adults who recognize themselves in their child's diagnosis. The female phenotype that we cover in our signs of autism in girls post overlaps significantly with this group.

Group 2: Mixed ASD with Developmental Delay (second-largest)

Roughly 25 to 30% of the studied population:

• Autism plus developmental delay across multiple domains
• Earlier age at diagnosis (often by age 3)
• More frequent language delay or difficulty
• Mixed cognitive profile, sometimes including intellectual disability
• Often identified through pediatric developmental screening
• Higher rates of certain medical co-occurrences (epilepsy, GI issues)

Group 3: Moderate Challenges

Roughly 20% of the studied population:

• Moderate level of autism characteristics across all domains
• Average or near-average cognitive development
• Diagnosis often in early childhood (3 to 5 years)
• Lower rates of severe co-occurring conditions
• More variable developmental trajectory; supports often produce noticeable progress

Group 4: Broadly Affected (smallest group)

Roughly 10 to 15% of the studied population:

• Significant challenges across communication, behavior, cognitive, and adaptive domains
• Higher rates of intellectual disability
• Often nonspeaking or minimally speaking
• Higher rates of medical co-occurrences (epilepsy, GI, sleep, motor)
• Earliest age at diagnosis on average
• Higher long-term support needs

This group is sometimes called "profound autism" in other research framings, though the terminology is contested. Our nonverbal autism resources cover the communication side that's most relevant to this group.

---

What This Means for Your Specific Child

Three things are worth saying directly:

You can't reliably assign your child to a subtype from this post. The subtypes were identified using detailed behavioral assessments combined with genetic data. Your developmental pediatrician's evaluation places your child on the autism spectrum and assigns a level designation; that's the relevant clinical information today. The research subtypes are a parallel framework that may eventually be assignable in clinical settings, but isn't yet.

Reading the subtype descriptions and recognizing your child is normal. Most parents will see their child somewhere in one of the four groups. That recognition is real, but it's not the same as a clinical assignment. Many children straddle two groups or share features across them.

Today's care is based on individual assessment, not subtype. Your child's IEP, therapy schedule, accommodations, and support needs are determined by what they specifically need, not by which research subtype they may eventually map to. The work of supporting your child is the same whether they're Group 1 or Group 3.

What might change in the next 5 to 10 years if subtype research holds up:

• Therapy recommendations may become more matched to subtype (some interventions may help Group 1 but not Group 4, and vice versa)
• Prognosis information may become more accurate and less generic
• Genetic counseling may become more useful for families considering additional children
• Drug research may target specific mechanisms for specific subtypes

None of this is currently happening in clinical care. Reading the research is informative; basing decisions on it isn't yet warranted.

---

How Subtypes Relate to the Level System

A common confusion: "Is my child Level 2 or Group 1?"

These are different frameworks answering different questions.

Level designation (Level 1, 2, or 3) describes how much support a child currently needs across two specific domains (social communication, restricted/repetitive behaviors). It's based on observable behavior at a point in time and can change as the child develops or environments shift. Levels are used in clinical practice to inform service decisions, IEP placement, and insurance authorization.

Subtype (Groups 1 through 4 in this research) describes what kind of autism a person has, based on combined behavioral, developmental, and biological data. Subtypes are presumed more stable because they reflect underlying biology rather than current support needs.

A child can be Level 2 in current support needs and Group 1 in subtype, or Level 2 and Group 3, or any combination. The two frameworks aren't competing; they describe different layers of the same child. Our autism levels explained post covers the level system in detail if you want the practical clinical framework.

---

Should You Pursue Genetic Testing?

The 2025 subtype research relies heavily on genetic data, which raises a question many parents have: should we get genetic testing for our child?

The current consensus among developmental pediatricians:

Genetic testing is recommended when autism is paired with:

• Intellectual disability or significant developmental delay
• Dysmorphic features or unusual physical findings
• Family history of autism, intellectual disability, or related conditions
• Specific neurological signs (seizures, microcephaly, macrocephaly)

Genetic testing is reasonable in other cases, depending on family interest, insurance coverage, and the developmental pediatrician's recommendation.

What testing does:

• Microarray (chromosomal microarray) identifies large genetic differences. About 10 to 15% of autism cases have a finding.
• Whole-exome sequencing looks at all protein-coding genes. About 25 to 40% of cases have a finding when this is added.
• Fragile X testing is sometimes done as a separate test (one of the most common genetic causes of autism).

What testing doesn't do:

• It doesn't tell you which "subtype" your child fits into in the 2025 research framework.
• It doesn't predict how your child will develop with high accuracy.
• It doesn't usually change current treatment, though specific genetic findings can rule certain conditions in or out.
• Most autism is polygenic (many small genetic differences combining), so most tests don't find a single causal variant even when they help.

If genetic testing is something you're considering, talk to your developmental pediatrician or ask for a referral to a pediatric geneticist. Insurance increasingly covers it when ordered with appropriate clinical justification.

---

What to Read and What to Do With This Research

For staying informed without overclaiming:

• The Simons Foundation publishes accessible summaries of their research at sfari.org
• SPARK (the dataset used in this work) is at sparkforautism.org and accepts new participants if you want to contribute data
• The 2025 paper is available on bioRxiv and in clinical journals; preprints are free to read

For practical purposes:

• Your child's clinical care doesn't change based on this research
• The IEP, therapy, accommodations, and support decisions you make are based on your child's individual needs
• Reading the research can help you understand why autism varies so much and why generic answers ("autistic kids do X") often don't fit your specific child
• If genetic testing is something you've been on the fence about, this research is one more reason to consider it, particularly if other indications apply

If you're sitting with a recent diagnosis and trying to figure out where to start, Beacon is a tool worth knowing about. It's an AI companion built specifically for autism parenting, available at the hours when nothing else is. Useful when you're trying to make sense of where your specific child fits, what to prioritize first, and what to leave for later, with something trained on autism instead of guessing what to Google next.

---

What This Means for Babies and Very Young Children

A common parent question after reading subtype research: can I tell which subtype my baby is?

The honest answer is no, not yet. The four subtypes were identified using a combination of behavioral data, developmental history, and genetic data. Most of that combined data isn't available for babies. A 9-month-old who's showing some early autism signs hasn't yet generated enough developmental history to be placed reliably in one of the four groups, and most haven't had genetic testing done.

A few patterns do emerge from the research that are worth knowing:

• The Group 4 (broadly affected) subtype tends to be diagnosed earliest, often by 18 to 24 months, because the signs are visible and pronounced
• The Group 1 (social/behavioral challenges) subtype is most often diagnosed late (after age 5, sometimes much later) because early signs are subtler
• Group 2 (mixed ASD with developmental delay) is also typically diagnosed early because pediatric developmental screening flags the delays

What this means in practice for parents of babies and very young children: the subtype framework is informative for understanding why autism presentations vary, but it doesn't change what to do for your specific baby right now. Early intervention services work across all four subtypes. The decision to refer to early intervention, push for evaluation, or seek a developmental specialist is the same whether your child eventually maps to Group 1 or Group 4.

For the practical baby and toddler signs you can look for now, see our signs of autism in babies post (covering 0 to 18 months) and signs of autism in toddlers (covering 6 months to 3 years).

---

A Note on Where the Field Is Going

The autism research landscape is in an unusual moment. Large datasets (SPARK, AGRE, others) combined with machine learning approaches are starting to produce findings that older small-sample studies couldn't. The 2025 subtype work is one example. Other lines of research are looking at:

• Brain imaging signatures across autism subtypes
• Sex-specific genetic and behavioral patterns (much of which informed our signs of autism in girls post)
• The relationship between autism and other neurodevelopmental conditions (ADHD, intellectual disability, language disorders)
• Long-term outcome studies across decades

The pace of progress is faster than it has been in 20 years, which means clinical care will likely look different by the time today's diagnosed kids are teenagers. That's reason for cautious optimism, not for changing decisions today based on speculative future treatments.

---

Where to Go Next

For the practical clinical framework most clinicians use today, see autism levels explained. For the broad signs picture, see signs of autism in toddlers. For what to do after a recent diagnosis, see what to do after autism diagnosis.

The subtype framework is research, not yet clinical reality. The work of supporting your child is the same whether they map to Group 1 or Group 4: meet them where they are, get the services that fit their needs today, and let the research evolve in the background. Your kid is the same kid no matter how researchers eventually carve up the spectrum, and they need you doing the practical work, not waiting for the science to settle.