Best free AI coding tools (2026): 10 options compared

The best free AI coding tools in October 2026 are GitHub Copilot Free and Cursor Free for editor autocomplete, Codex (included in ChatGPT Free) and Google Antigravity for a hosted agent at $0, and opencode, Aider, Cline or Qwen Code on a local Ollama model if you want no usage cap at all. Every hosted free tier has a limit, and two well-known free tiers (Gemini CLI’s personal sign-in and Qwen OAuth) closed this year, so check the details below before you build a workflow on one.

This list covers ten tools with a real free plan or free use. Every limit and price was checked on the vendor’s official page on October 28, 2026. Where a vendor does not publish a number, we say so rather than guess.

Quick comparison

Most of these are covered in more depth in the best AI coding tools, which lines up 18 CLI agents by sign-in, MCP and effort levels.

Hosted tools with a free plan

GitHub Copilot Free

GitHub’s free plan is the lowest-friction start: sign in with a GitHub account and Copilot works in VS Code, JetBrains IDEs and the terminal. GitHub’s plan docs say Copilot Free is “limited to 2000 completions per month” and comes with “an allowance of GitHub AI Credits” for chat and agent use. The credit amount isn’t listed on the plans page. On Free, models are picked through auto model selection only.

Copilot CLI is included in every plan, Free too:

npm install -g @github/copilot
copilot

Limits: no cloud agent that opens pull requests for you, no Copilot code review and no custom instructions on Free. Heavy agent use runs through the credit allowance fast. Paid: Pro is $10 a month and Pro+ $39 a month as of October 2026. Verified students get the Copilot Student plan for free, and verified teachers and maintainers of popular public projects may get Pro for free. Fits: anyone who wants completions and occasional chat without a card. See how to use Copilot CLI for the terminal side.

Codex with a ChatGPT Free account

OpenAI includes Codex in ChatGPT Free, Go, Plus, Pro, Business, Edu and Enterprise. Its pricing page describes the Free tier as a way to “explore Codex capabilities on quick coding tasks” and publishes no exact limits for it.

npm install -g @openai/codex
codex          # choose "Sign in with ChatGPT"

Strengths: a full terminal agent with a real OS sandbox, AGENTS.md, MCP and codex exec for scripts. Limits: the Free allowance is small and unpublished, and model choice is restricted. Free fallback: Codex can also run against a local model with codex --oss (Ollama or LM Studio), which has no quota at all. Fits: trying an agentic CLI before paying for anything. How to use Codex CLI walks through the setup.

Google Antigravity (and what happened to Gemini CLI’s free tier)

For two years, Gemini CLI’s Google sign-in was the most generous free tier around. That changed. Google’s notice says that on June 18, 2026, unpaid tier and Google One users were moved from Gemini CLI to Antigravity CLI. Gemini CLI keeps working with paid Gemini API keys, Vertex AI and Gemini Code Assist Standard or Enterprise licenses, so it is no longer a free option for individuals.

The free route from Google is now Antigravity. Its pricing page lists an Individuals plan at $0 a month with “a variety of agent models”, unlimited Tab completions, unlimited Command requests and “basic weekly rate limits”. Google AI Pro and Ultra subscribers get higher limits and a credit pool. The CLI runs as agy:

# macOS, Linux
curl -fsSL https://antigravity.google/cli/install.sh | bash
# Windows PowerShell
irm https://antigravity.google/cli/install.ps1 | iex

Strengths: a desktop app for macOS, Windows and Linux plus a CLI, and agy plugin import gemini converts Gemini CLI extensions. Limits: weekly caps without published numbers, and Google says there’s no one-to-one feature parity with Gemini CLI. Fits: former Gemini CLI free users. If you still run Gemini CLI on a paid key, how to use Gemini CLI covers the June change.

Cursor Free

Cursor’s pricing now lists Free, Plus ($8 a month), Premium ($30), Super ($100), Ultra ($200) and Teams. Cursor’s help center says Free gives “access to Cursor’s core features with limited usage”: Agent, Chat and Tab completions with Cursor’s own models (Grok and Composer). The old Hobby plan works the same way. Cursor runs on macOS, Windows and Linux, and its CLI installs with:

curl https://cursor.com/install -fsS | bash

Limits: usage caps are not published as numbers, and Bugbot code review starts at Premium. Fits: developers who want to try an AI-first editor before choosing a plan. If Cursor isn’t the right fit, see the best Cursor alternatives.

Devin Desktop (formerly Windsurf)

Windsurf was renamed Devin Desktop on June 2, 2026, and windsurf.com now redirects to Devin’s pricing page. Cognition says existing users keep their plan. The free plan lists a “light quota to code with agents”, “limited model availability”, unlimited inline edits and unlimited Tab completions. Pro is $20 a month and Max $200 a month as of October 2026. Devin Desktop can also manage third-party agents through the Agent Client Protocol.

Fits: people who mainly want unlimited autocomplete and a small amount of agent work. Limits: the agent quota is deliberately light, and the model list on Free is restricted.

Kiro

Kiro, from AWS, has an IDE and a CLI (kiro-cli). Its pricing page lists the free tier at 50 credits with access to open-weight models and Claude Sonnet 4.5. Free users who sign up with social login or AWS Builder ID also get Qwen3 Coder Next, DeepSeek 3.2 and MiniMax M2.1, subject to rate limits. Paid plans are Pro ($20, 1,000 credits), Pro+ ($40), Pro Max ($100) and Power ($200). The IDE runs on macOS, Windows and Linux; the CLI on macOS, Windows 11 and Linux.

Fits: trying spec-driven development, or open-weight models without running them locally. Limits: 50 credits go quickly on agent tasks.

Free tools you bring a model to

These four don’t sell you a model plan you have to buy. The tool is free; you point it at a provider’s free models, a paid API key or a model running on your own machine. If you want the full local setup, local LLM coding agent covers Ollama, LM Studio, context length and hardware.

One setting matters for all of them: raise Ollama’s context length, because it silently drops what doesn’t fit.

ollama pull <model>
OLLAMA_CONTEXT_LENGTH=64000 ollama serve

opencode

opencode is a terminal agent that works with many providers. Its hosted catalog, OpenCode Zen, lists several free models as of October 2026, most of them free “for a limited time”. Read the notes: for some, data collected during the free period may be used to improve the model, and the Nemotron models are marked trial use only, with a warning not to send personal or confidential data. Set OPENCODE_DISABLE_FREE_MODELS=1 to hide them.

For a fully local setup, add Ollama as a provider in opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "ollama": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Ollama (local)",
      "options": { "baseURL": "http://localhost:11434/v1" },
      "models": { "qwen3": { "name": "Qwen3 (local)" } }
    }
  }
}

Fits: a free daily-driver CLI that can switch between free, paid and local models. Details in how to use opencode.

Aider

Aider is a free pair programmer for the terminal that commits each change to git. It talks to Ollama directly:

python -m pip install aider-install
aider-install
export OLLAMA_API_BASE=http://127.0.0.1:11434
aider --model ollama_chat/<model>

Aider’s docs recommend the ollama_chat/ prefix and warn that Ollama silently discards context beyond its window; by default Aider sizes the window per request. Fits: developers who want small, reviewable commits. Limits: it doesn’t take MCP servers. See how to use Aider.

Cline

Cline’s pricing page says the extension is “completely free for individual developers”; you pay only for AI inference when you use a paid model. To run it free, choose Ollama as the provider in Cline’s settings, set the base URL to http://localhost:11434 and pick your model. Cline’s docs suggest turning on Use Compact Prompt and keeping tasks small for local models. There is also a cline CLI. Fits: VS Code users who want an agent in the editor without a subscription.

Qwen Code

Qwen Code is a free terminal agent. One thing changed this year: its docs say the Qwen OAuth free tier was discontinued on April 15, 2026, so new requests through it are rejected. The CLI still supports OpenAI, Anthropic, Gemini and Qwen APIs, any third-party provider and local models through Ollama or vLLM:

npm install -g @qwen-code/qwen-code@latest
export OPENAI_BASE_URL=http://localhost:11434/v1
export OPENAI_API_KEY=ollama
export OPENAI_MODEL=<model>
qwen

Fits: local-first users who like the Qwen Code workflow. Limits: without OAuth there’s no free hosted model anymore.

What about Claude Code?

Claude Code is not free. The CLI installs at no cost, but the Free Claude plan doesn’t include it; you need Pro, Max, Team, Enterprise or API credits. Claude Code pricing compares the plans, and the best Claude Code alternatives covers tools to use alongside or instead of it.

How to choose

• You want autocomplete in your editor: Copilot Free (2,000 completions a month), Devin Desktop (unlimited Tab) or Cursor Free.
• You want an agent in the terminal, hosted, for $0: Codex with ChatGPT Free, or Antigravity’s agy. Expect weekly or unpublished caps.
• You want no cap at all: opencode, Aider, Cline or Qwen Code on Ollama. Budget for hardware and smaller models.
• You care about where code goes: a local model keeps prompts on your machine. Be careful with free hosted models whose terms allow training on free-period data.
• You hit limits often: combine tools. Use a free hosted tier for hard problems and a local model for routine edits, tests and docs.

A practical pattern is to stack free tiers: Copilot Free in the editor, a hosted free agent for the occasional big task, and a local agent for everything repetitive. When one runs out, you switch rather than wait.

Where AgentVera fits

AgentVera is a desktop app for macOS, Windows and Linux that runs coding CLIs side by side in real terminals. It doesn’t replace any tool above and has no model of its own. It is the place to run several of them at once, which is how you get the most out of overlapping free tiers. Its free plan has unlimited agents, and each CLI keeps its own login or local model, so the free tiers above apply unchanged.

• Workspace: a grid of agents per project, from 1×1 to 5×4. Put Codex on ChatGPT Free next to opencode on Ollama and Copilot CLI in the same project.
• Worktrees: start each agent in its own git worktree from the Git panel, so two agents never edit the same files.
• Code review: have one agent review what another changed before you merge, useful when a local model wrote the code.
• MCP manager: set up MCP servers once; they attach to supported CLIs on every launch.
• Token savings: trims what agents send, which stretches a capped free tier.

The grid, worktrees, code review, the MCP manager and token savings are on the free plan; Pro and Team add flows, night shift, auto approval and automatic account switching. If you don’t need several agents at once, a single CLI in a terminal works fine. For other ways to run agents in parallel, see the alternatives pages. To try it, download AgentVera or compare plans.

Wrapping up

Free AI coding is real in 2026, but it moved. Gemini CLI’s personal free tier became Antigravity, Qwen OAuth closed, Windsurf became Devin Desktop and Cursor reshaped its plans. Start with Copilot Free or Cursor Free in the editor, Codex or Antigravity for a hosted agent, and opencode or Aider on Ollama when you want no cap. Recheck the vendor pages before you rely on a limit, since free tiers change more often than paid ones.

More roundups

• Best Claude Code GUI apps (2026)
• Best Cursor alternatives in 2026: 10 tools compared
• Best Claude Code alternatives in 2026: 9 CLI agents