---
title: "Frontier AI agents · the general-purpose flagships · Agent Fieldbook"
url: https://agentfieldbook.org/catalog/frontier/
description: "Frontier agents: the general-purpose flagship assistants at the leading edge of what each lab's models can do. The reference point the rest of the catalogue is built on and compared against, each specced against primary sources and drawn as a topology specimen."
section: "Catalogue · Frontier"
source: Agent Fieldbook — generated from the published page
---

# Frontier agents

The general-purpose flagships: the broad, capable assistants each lab ships at the leading edge of its models. They set the capability floor and ceiling for everything else in the catalogue, because most copilots, platforms and vertical agents wrap a frontier model. Each card shows its topology, country and build mode; click through for the full spec.

## 7 agents Live data · click a specimen for its anatomy

## Field notes

The general-purpose flagships the rest of the catalogue is built on

The frontier tier is narrow: seven general-purpose flagships against 96 specialised agents. Nearly all are built by labs that also sell the underlying model, so the same organisation owns the reasoning core and the assistant wrapped around it.

That vertical integration is the pattern worth tracking. It gives the frontier labs first use of each capability and sets the reference point every specialised agent is measured against. As the model layer commoditises, the differentiation moves to what the flagship can reach, its tools, memory and connectors, rather than to raw reasoning.

- [Brandon Chaplin](https://www.linkedin.com/in/brandon-chaplin-digital-marketing-strategist)

## Common questions

### What is a frontier AI agent?

The general-purpose flagship assistant a lab ships on its newest model. It is built for open-ended work across any subject rather than one industry, and it is usually the system new benchmark results are quoted against.

### What is the difference between a frontier model and a frontier agent?

The model does the reasoning. The agent is the model plus the scaffolding around it: planning, tool use, memory and the ability to run a task over several steps. API access to a model does not give you that behaviour on its own.

### Which frontier AI agent is the best?

It depends on the task, and the ranking changes with every release. Coding, long-document work and tool use are strengths that vary between labs. Test two or three on your own work; a headline benchmark score is a starting point, not an answer.

### Why are there so few frontier agents?

Because training a frontier model costs a fortune and only a handful of organisations can fund it. Nearly every frontier agent is built by the same lab that trains the model underneath, which is why the tier stays small while everything else builds on top of it.

## Next in the learning path

- [The full catalogue Every agent, not just the frontier tier](https://agentfieldbook.org/catalog/all/)

- [Copilots & embedded Frontier models scoped into a product](https://agentfieldbook.org/catalog/copilots_embedded/)

- [Agent platforms The substrates frontier agents run on](https://agentfieldbook.org/catalog/platforms/)

- [Agent topologies The shapes these flagships coordinate through](https://agentfieldbook.org/design/topologies/)

## Entries

_7 entries listed on this page._

| name | description | subjectOf |
| --- | --- | --- |
| ChatGPT Agent | Distribution is the moat, destination product and launch surface for new OpenAI agent capabilities. | https://openai.com/index/march-funding-updates/ |
| Claude Agent | Not only a model destination, Anthropic shaped the agent tooling layer through MCP. | https://www.anthropic.com/system-cards |
| Gemini Agent | Combines product distribution, model ownership, and a standards play through A2A. | https://blog.google/innovation-and-ai/sundar-pichai-io-2026/ |
| Mistral Vibe | Rebranded from Le Chat to Vibe (Jun 2026); EU-sovereign frontier assistant now runs Mistral Large 3 / Medium 3.5. |  |
| OpenAI Deep Research | ChatGPT mode that plans + browses + synthesises multi-source research reports over 5-30 minutes. | https://openai.com/index/introducing-deep-research/ |
| Gemini Deep Research | Gemini Advanced's multi-step research agent: plans, browses, and synthesises long-form reports. | https://blog.google/products/gemini/google-gemini-deep-research/ |
| Perplexity Deep Research | Perplexity's deep-research mode, plans, browses, and synthesises reports with citation trail. | https://www.perplexity.ai/hub/blog/introducing-perplexity-deep-research |