---
title: "MCP, GDPval, SWE-bench: 8 AI Terms Defined"
url: https://nativefirst.ai/blog/definitions/
description: "MCP, GDPval, SWE-bench, open weights and four more. The eight terms every AI vendor deck assumes you already know, defined for the buyer."
author: John Tan
category: "Definitions"
publisher: nativefirst.ai
format: Markdown copy of the web page at the url above, same text.
---

# 8 AI terms, explained plainly.
Starting with MCP, GDPval and SWE-bench.

These eight terms appear in almost every AI vendor deck and model launch. Each one below has a short answer and a link to the full explainer, written for the person signing the invoice rather than the person writing the integration. Start anywhere.

11

Terms

4

Groups

50m

To read all of it

Agents 4 terms

What people mean by “agent”, and what each level actually commits you to.

[01 What Is a Level-3 AI Agent? A Level-3 agent completes a whole task end to end and reports back, instead of answering a question or following one instruction. Most companies are running Level 1 and calling it Level 3.](https://nativefirst.ai/blog/what-is-a-level-3-ai-agent/) [02 What Is an Agent Teammate? An agent teammate owns a task, decides inside boundaries you set, and reports back. A tool waits to be asked. The difference is who holds the work.](https://nativefirst.ai/blog/what-is-an-agent-teammate/) [03 What Is Instinct? Instinct is a consumer AI agent that runs entirely inside iMessage, with its own phone number and computer. It raised $1 billion at a $10 billion valuation in September 2026 and still has no mobile app.](https://nativefirst.ai/blog/what-is-instinct/) [04 What Is Muse? Muse is Meta’s free personal AI agent. It tops the community Assistant Benchmark at 9.1, though on 7 of the 15 dimensions against Instinct’s 11, so it leads on a narrower evaluation.](https://nativefirst.ai/blog/what-is-muse/)

Infrastructure 2 terms

How agents connect to your systems, and what has to exist before they can.

[01 What Is an MCP Server? MCP, or Model Context Protocol, is a standard way to connect an AI agent to your live internal systems. Without it an agent is cut off from the data that would make it useful.](https://nativefirst.ai/blog/what-is-an-mcp-server/) [02 What Is an Agent Operating System? An Agent OS is the shared layer that gives every agent in a company the same context, rules and permissions. Without one, each agent starts from scratch knowing nothing the others know.](https://nativefirst.ai/blog/what-is-an-agent-operating-system/)

Benchmarks 3 terms

The three scores every model launch quotes. What each one measures, and what it leaves out.

[01 What Is GDPval? GDPval is OpenAI’s benchmark that scores AI on real work drawn from 44 occupations, graded by working professionals. It measures finished deliverables rather than exam answers.](https://nativefirst.ai/blog/what-is-gdpval/) [02 What Is SWE-bench? SWE-bench measures whether an AI can fix real GitHub issues so the existing tests pass. Its Verified split was retired over contamination, so SWE-bench Pro is the number that counts now.](https://nativefirst.ai/blog/what-is-swe-bench/) [03 What Is the Artificial Analysis Intelligence Index? The Artificial Analysis Intelligence Index is the composite score behind most “smartest model” headlines. It combines several evaluations into a single number, which is both its use and its limit.](https://nativefirst.ai/blog/what-is-the-aa-intelligence-index/)

Models 2 term

What you are buying when you buy access to a model, and what you get to keep.

[01 What Are Open Weights? Open weights means the model file itself is published, so you can run it on your own hardware for as long as you like. It is not the same thing as open source.](https://nativefirst.ai/blog/what-are-open-weights/) [02 What Is Jev? A System One model returns typed judgments with calibrated probabilities instead of text. Jev is the first one. It answers with a choice, a score or a probability, and never with a sentence.](https://nativefirst.ai/blog/what-is-jev/)

Going further Three places to go next →

Where to go once you have the vocabulary.

[→ The Benchmark Index Nineteen benchmark posts since November 2025, in one place. Four of those eleven months contained a reset that invalidated earlier numbers.](https://nativefirst.ai/blog/the-benchmark-index/) [→ The Operator Guides One guide per frontier model. What it costs, what to route to it, how to govern it, and when to say no.](https://nativefirst.ai/blog/guides/) [→ The Playbook Everything else. Written for founders and CEOs deciding where the AI budget goes this quarter.](https://nativefirst.ai/blog/)
