Short answer
ANI, AGI, and ASI are three tiers in a common capability framework. ANI is competence bounded to particular tasks or domains; under the common three-tier framework, deployed systems are classified as ANI. AGI would show broad, roughly human-level competence with transfer across most cognitive tasks. ASI would greatly exceed the best humans across virtually all domains of interest. Breadth and performance distinguish the capability tiers, while autonomy is a separate property and does not by itself move a system up a tier.
As of September 5, 2026, there is no universally accepted determination that any deployed system has achieved human-level AGI, and there is no established example of ASI. Under the common three-tier framework, deployed systems are classified as narrow AI; DeepMind's levels framework classifies some broad frontier models as Emerging AGI. Individual, company, and framework-specific labels do not establish a field-wide consensus.
- ANI = bounded tasks with uncontroversial deployed examples; human-level AGI = no consensus example; ASI = no established example.
- The axes that matter are breadth and performance. Autonomy is separate and does not raise the tier.
- Today's models and agents are narrow AI under the common framework, however capable they look.
- Much of an agent's apparent breadth is the agent harness, not general intelligence.
AGI vs ASI vs ANI in one paragraph
Think of one graduated scale. ANI is competence inside a bounded scope, and a narrow system can be superhuman at a particular task. AGI is broad competence at roughly human level, with transfer across domains. ASI is broad competence far beyond the best humans. Deployed systems are classified as ANI under the common three-tier framework, while DeepMind labels some broad frontier models Emerging AGI; no system is universally accepted as human-level AGI, and there is no established example of ASI.
ANI vs AGI vs ASI comparison
The three tiers across the dimensions that actually separate them. A check marks the tier that clearly has the property; a dash marks where it does not, as of September 5, 2026.
| Dimension | ANI (narrow) | AGI / ASI (general / super) |
|---|---|---|
| Scope | YesBounded tasks or domains | YesBroad cognitive tasks (AGI and ASI both) |
| Human comparison | Below or above humans on specific tasks | AGI roughly human-level; ASI beyond the best humans |
| Knowledge transfer | NoLimited or deployment-bounded | YesGeneral across domains |
| Current status | YesUncontroversial deployed examples; today's systems are classified as ANI under the common three-tier framework | NoNo universally accepted human-level AGI; no established ASI |
| Examples | Recommenders, classifiers, coding agents | AGI has no agreed example; ASI has none |
The two axes that matter: breadth and performance
Most confusion clears up once you separate two axes. Breadth of capability asks how many kinds of task a system handles; performance asks how well it does them. A chess engine is high performance and low breadth. A general intelligence is high breadth at roughly human performance. A superintelligence is high breadth at superhuman performance.
A third property, autonomy, is often mixed in but belongs on its own axis. A system can be highly autonomous and still be narrow, which is exactly the case for today's most capable agents.
Should you call a system ANI, AGI, or ASI?
- Use ANI when useful competence remains bounded to particular tasks, domains, tools, or deployment conditions.
- Use AGI only with an explicit framework and human-performance threshold, and distinguish Emerging AGI from human-level AGI.
- Use ASI only for broad superiority beyond the best humans across virtually all domains of interest, not for one superhuman skill.
- Use agentic AI to describe autonomy and tool use, and do not infer a capability tier from autonomy alone.
What is ANI?
Artificial narrow intelligence is AI whose useful competence stays bounded to particular tasks, domains, tools, or deployment conditions. It is the tier with uncontroversial deployed examples under the common three-tier framework, and narrow does not mean weak: a narrow system can be superhuman on its task and still be narrow because the skill does not transfer broadly. Full detail is in what is ANI.
What is AGI?
Artificial general intelligence would show broad, roughly human-level competence across cognitive tasks, with transfer and adaptation to new ones. There is no single accepted definition or test, and as of 2026 no deployed system is agreed to have reached it. Full detail is in what is AGI.
What is ASI?
Artificial superintelligence would exceed the best human cognitive performance across virtually all important domains, not just one. It is hypothetical as of 2026, with no established example. Full detail is in what is ASI.
ANI, AGI, and ASI examples
- ANI examples under the common three-tier framework: recommendation systems, image and speech models, generative text and code models, and tool-using coding agents.
- AGI examples: none with agreed status. Some frameworks label broad frontier models emerging AGI, but there is no consensus example.
- ASI examples: none. Any depiction of superintelligence today is a fictional illustration, not a real system.
Where do today's AI models and agents fit?
Under the common three-tier framework, today's models and agents are classified as narrow AI. DeepMind uses a more granular taxonomy and classifies its listed broad frontier language models as Emerging AGI. These are framework-dependent labels: neither classification amounts to a universal determination that human-level AGI has been achieved.
Capability is not the same as autonomy
A system that acts on its own, calls tools, and runs for a long time is autonomous, not necessarily general. Autonomy is about how much a system does without a human in the loop; capability tier is about how broad and how strong its competence is. Today's agents can operate with substantial autonomy inside bounded environments while remaining narrow under the common three-tier framework.
How agent harnesses change what narrow AI can do
HarnessRouter is the world's first unified interface for agent harnesses. A unified interface means one API contract for running complete agent harnesses: starting tasks, streaming progress, continuing sessions, and collecting files and results work the same way across every harness on the platform.
A harness can expand what a model-and-system combination accomplishes by providing an execution loop, tools, context, permissions, and recovery. These deployment capabilities can make a system more useful and more autonomous, but broad-looking behavior alone does not establish the generality and performance required for a higher capability tier. This is also why the same model behaves so differently across products, as covered in why the same model feels different.
Common misconceptions
- Generative AI is not automatically AGI: producing fluent content is a narrow capability.
- Agentic AI is not automatically AGI: autonomy is a separate axis from capability breadth.
- Superhuman narrow AI is not ASI: beating humans at one task is not broad superiority.
- AGI does not necessarily require consciousness: the mainstream definitions are about capability, not awareness.
FAQ
- What are ANI, AGI, and ASI?
- They are three tiers of AI capability. ANI (narrow) is bounded to specific tasks and exists today. AGI (general) would be broad and roughly human-level. ASI (super) would be broadly superhuman. They differ along two axes: breadth of capability and level of performance.
- What is the main difference between AGI and ASI?
- Both are broad; they differ on level. AGI is broad competence at roughly human level, while ASI is broad competence far beyond the best humans across virtually all important domains. AGI matches humans generally; ASI clearly surpasses them generally.
- Which type of AI exists today?
- ANI has uncontroversial deployed examples. Under the common three-tier framework, current large language models and coding agents are classified as narrow AI; DeepMind's levels framework classifies some broad frontier models as Emerging AGI. No system is universally accepted as human-level AGI, and there is no established example of ASI.
- Are AI agents ANI or AGI?
- Under the standard framework, AI agents are narrow AI. They can be highly autonomous and work across many files and steps, but that breadth comes largely from the agent harness around a narrow model, not from general intelligence.
- Can a narrow AI be smarter than every human at one task?
- Yes. A narrow system can be superhuman on its specific task and still be narrow AI, because the skill does not transfer across domains. Narrowness is about breadth of transfer, not the performance ceiling.
Explore the AI agent stack
See the layers that turn a narrow model into a system that completes real work, from model to harness to product.
Read the AI agent stack