Field NoteCoding / Builder Tools

Claude Code vs Codex vs Antigravity vs T3

Four tools, four approaches to model access and openness. Claude Code is locked to Claude and closed source. Codex is locked to GPT but the CLI is fully open source. Antigravity runs on a Google login yet lets you switch between Gemini, Claude, and open models. T3 is a cheap multi-model chat. Here are the facts — open-source level, model lock-in, cost, and how you interact with each.

➔ Operator Field Assessment

💡

What It Is

Four tools for AI-assisted coding — terminal agents (Claude Code, Codex), an agent-first IDE (Antigravity), and a multi-model web chat (T3) — each with different model access, integration depth, and autonomy.

🔓

Open Source

Codex CLI is fully open source (Apache 2.0). Claude Code is proprietary/source-available. Antigravity is closed source (it replaced the open Gemini CLI). T3 Chat is closed, but has a large open-source clone ecosystem.

🔌

Where It Fits

Developer IDE setup, agent orchestration, rapid prototyping, and daily coding workflows.

💳

Subscription vs. API

A personal subscription login keeps casual costs low (and as of 06/2026 these subs are heavily subsidized). API access bills per token — at enterprise scale that drives costs up fast, so budget for it.

🛠️

What I Use It For

Refactoring legacy codebases, scaffolding web apps fast, writing complex automations, and comparing model outputs on the same problem.

🛡️

Risk & Governance

Terminal and agent tools can execute shell commands — restrict permissions to project folders only. Never run in directories with credentials or root access.

➔ Tool-by-Tool Breakdown

Toggle between tools to compare open-source level, model access, pricing, and dev sentiment.

Claude Code

Anthropic

Terminal-native agentic coding assistant by Anthropic

Open SourceProprietary

Source-available — you can read the code but cannot fork, redistribute, or self-host it.

ModelsClaude only

Locked to Anthropic models. No provider switching.

Subscription vs. API

Subscription login

Pro ~$20/mo · Max ~$100–200/mo

API access

Pay-per-token via Anthropic API

Claude Code left the bundled Pro plan in April 2026; heavy usage moved to metered credit pools in June 2026.

How You Interact

Terminal CLIRun `claude` in any project directory. Reads files, runs commands, edits code in autonomous loops.
VS Code ExtensionSidebar panel + inline edits. Same agent capability without leaving the editor.
JetBrains / CursorAvailable as a plugin for IntelliJ, WebStorm, and Cursor.

Available As

Terminal CLIVS Code extensionJetBrains extension

Strengths (dev sentiment)

  • +Highest code quality in blind tests (~67% win rate vs. Codex)
  • +Full MCP (Model Context Protocol) support — a differentiator vs. Codex
  • +Large context window handles big refactors and stays focused
  • +Strong on small, precise multi-file changes

Limits (dev sentiment)

  • −Usage limits are the top complaint — one complex prompt can burn 50–70% of a 5-hour window
  • −Affordability concerns once you leave the subscription for API billing
  • −Locked to Claude models — no vendor flexibility
  • −Closed source — no self-hosting or inspection
Typically For

Complex reasoning, multi-file refactors, and agent loops that run commands

⚠ Subscription vs. API — the cost trap at scale

All four let you log in with a personal subscription, which keeps costs low — and as of June 2026 those consumer subscriptions are heavily subsidized. Inside an enterprise you typically run on API keys instead, which bill per token. A single agentic task can push hundreds of thousands to millions of tokens, so an API budget runs out far faster than a flat personal sub would suggest. Plan capacity around token volume, not seat count.

Agent vs. Autocomplete vs. Chat

These four fall into different categories. Claude Code and Codex CLI are terminal agents — they read your file system, run commands, and loop until the task is done. Google Antigravity is an agent-first IDE (a VS Code fork) where you delegate tasks to agents and pick which model runs them. GitHub Copilot, built on Codex, is an autocomplete engine that suggests the next line at your cursor. T3 Chat is a multi-model web chat — you paste code and it responds, with no file-system access.

Open Source: Where Each Tool Stands

The four span the full spectrum. Codex CLI is fully open source under Apache 2.0 (the CLI on GitHub — the GitHub Copilot extension is a separate, closed product). Claude Code is proprietary and source-available: you can read the code but cannot fork, redistribute, or self-host it. Google Antigravity is closed source; it is built on the open VS Code base but replaced the Apache-2.0 Gemini CLI, which Google is sunsetting. T3 Chat the product is closed source, but it spawned a large open-source clone ecosystem (the "Cloneathon"), with dozens of MIT-licensed forks you can self-host.

Model Lock-in

Claude Code runs only on Claude; Codex runs only on GPT. If a provider has an outage or changes pricing, a single-vendor tool is exposed. Antigravity is designed around the opposite approach — one Google login, with the ability to switch between Gemini, Claude, and open models per task. T3 takes a bring-your-own-key approach: connect any provider's key and route costs directly to that provider with no platform markup.

IDE and Editor Coverage

GitHub Copilot (Codex) has the widest editor reach — VS Code, JetBrains, Vim, Neovim, and Azure DevOps. Claude Code ships a VS Code and JetBrains extension that runs the full agent in a sidebar. Antigravity is itself an IDE — a VS Code fork — plus a CLI and SDK. T3 has no IDE extension; it is browser-only.