Bottom Line: Choose by access route and cost

Seven terminal clients can be useful at $0 software cost, but "free" can mean several different things:

a free or open-source client whose model calls are billed separately;
an account quota bundled with Google, ChatGPT, or GitHub access;
an unpaid API tier with its own model and request limits; or
a local model available without a license fee that still uses your hardware and electricity.

Google's consumer route changed: its transition announcement moved free Gemini Code Assist for individuals and Google AI Pro/Ultra users from Gemini CLI to Antigravity CLI on June 18, 2026. Do not choose Gemini CLI for the old 1,000-request daily offer. Check Antigravity's current plans instead.

Codex pricing currently describes limited Codex access on the Free plan, plan-specific ChatGPT usage, and separate API-key billing. The open clients in this guide do not bundle a universal model allowance. GitHub documents Copilot Free as a limited plan, while Copilot CLI usage is measured through the current AI-credit and plan system. The pages were checked on 2026-10-27, so use them again before buying or relying on a quota.

The right choice depends on how you obtain model access, whether inference runs through a provider or on your machine, and which permissions and workflow integrations you need. Use the comparisons below to match those conditions to your work.

For the wider market, see the AI CLI Tools Complete Guide.

What "Free" Covers

Meaning	What you get	What you still pay or manage
Free client	The terminal software can be installed without a license fee	Model provider, API calls, account limits, and updates
Included account quota	A provider includes some requests or credits with an account plan	Eligibility, rate limits, model availability, and changing plan terms
Unpaid API tier	A provider offers limited API calls without billing enabled	Smaller quotas, restricted models, and possible service limits
Local inference	No cloud inference bill for a local model	Hardware, storage, electricity, setup, and model quality
Promotion or trial	Temporary access under published terms	End date, eligibility, and the need for a fallback route

A local CLI can still send prompts to a hosted model. Only a local provider or model makes inference local; file and tool execution location is a separate choice. Keep the client, model route, and account plan as separate rows in your notes.

The Seven Tools at a Glance

Tool	Free part	Model or account route	Cost boundary	Useful when
Gemini CLI	Free client software, not the former consumer free quota	Supported Code Assist Standard/Enterprise or paid API access	Consumer accounts moved to Antigravity CLI; model access is separate	Your organization or paid API route still supports Gemini CLI
Codex CLI	The client can be used with the current Free plan's limited Codex access	ChatGPT sign-in or OpenAI API key	ChatGPT limits and API billing are separate; cloud requires ChatGPT sign-in	You already use ChatGPT or need a local OpenAI coding loop
OpenCode	Open-source client; selected limited-time free models	API key, OAuth/subscription, or local provider depending on route	Provider terms and charges or local hardware are separate	You want to choose providers and keep the client independent
goose	Open-source Apache 2.0 agent	Providers, subscriptions through ACP, or local routes	Model access follows the selected provider or local hardware	You need extensions, recipes, or agent workflows
aider	Terminal pair-programming client	Cloud providers or local models such as Ollama	Model calls and local hardware are outside the client	You prefer focused edits in a local Git repository
Crush	Open-source terminal agent	Select an LLM or use OpenAI- or Anthropic-compatible APIs	No universal model quota is bundled	You want a model-switchable TUI with LSP and MCP options
GitHub Copilot CLI	Copilot Free has limited access for eligible individuals	GitHub Copilot plan or your own provider key	Plan limits and AI credits apply; BYOK changes which GitHub features are available	Your work is centered on GitHub repositories, issues, and pull requests

This table describes access paths, not a rank. Check the linked official page for the current account, model, and region rules.

Provider-by-Provider Details

1. Gemini CLI: check the consumer transition first

The Gemini CLI documentation banner says unpaid-tier and Google One users moved to Antigravity CLI on June 18, 2026. The older table below it still displays 1,000/day and 250/day rows. Those rows alone are not evidence of a currently available consumer free route.

Individual users: follow the Antigravity quickstart and its own plan limits. The old Gemini CLI daily allowance does not transfer.
Supported enterprise and paid API users: Google's transition notice preserves Code Assist Standard/Enterprise and paid Gemini or Gemini Enterprise Agent Platform API access. Check your organization's setup and billing before using the Gemini CLI authentication guide.
Existing configurations: use the Antigravity migration guide; do not assume every command or setting is identical.

Choose Gemini CLI when: you have confirmed a continuing supported account or paid API path. For a new personal setup, start with Antigravity CLI and check its current eligibility and limits. Sources checked on 2026-10-27.

2. Codex CLI: local client, plan limits, and API billing

Codex CLI's official guide starts in a local project directory. OpenAI's current pricing page lists limited Codex access for Free, expanded access for paid ChatGPT plans, and an API-key path billed by API usage. The exact model and allowance depend on the current plan and account.

For hosted models, choose between plan-dependent ChatGPT access and separately billed API usage. Codex Cloud requires ChatGPT sign-in. (Codex CLI; pricing; authentication, checked 2026-10-27.)

Choose this route when: you already have an eligible ChatGPT plan, want to use an OpenAI API key for automation, or need a local CLI that can optionally hand work to a prepared cloud environment. Check /status and the plan page for your current allowance. For local inference, the CLI reference documents --oss with Ollama or LM Studio; that route depends on your local model and hardware, not the ChatGPT quota.

3. OpenCode: provider choice, subscriptions, or local models

OpenCode's provider docs list 75+ providers and local models. The connection method depends on the provider: the docs describe API keys, OAuth/device login, and existing subscriptions. They list ChatGPT Plus, GitHub Copilot, and GitLab Duo as zero-setup subscription routes, and document local Ollama, LM Studio, and llama.cpp routes. Anthropic's section explicitly prohibits Claude Pro/Max plugins, so do not assume every subscription transfers.

OpenCode itself is a free client; model access can still be provider-funded, subscription-backed, or local. The provider page says NVIDIA's build.nvidia.com route offers free access to open models, but terms, limits, and model availability remain provider-specific. The current OpenCode Console model catalog also lists selected limited-time free models alongside paid models. Its free-tier chat models require OpenCode, and data-retention or training terms vary by model. The paid route uses credits and an API key; check the selected model and its terms rather than treating every Console or Zen request as either free or paid.

4. goose: open source with provider and extension choices

The official goose repository links goose-docs.ai for its Quickstart, Installation, and Documentation. The repository describes an Apache 2.0 agent that runs on your machine with a desktop app, CLI, and API. It supports multiple providers, existing subscriptions through ACP, local routes such as Ollama, recipes, MCP extensions, and subagents. The agent process can run locally while inference is remote; only a local provider or model makes inference local.

Those options do not create one universal free tier. A subscription may cover model usage under that provider's terms; an API key may incur token charges; a local model shifts the cost to your machine. Choose goose for the workflow and extension surface, then verify the selected provider separately.

5. aider: pair programming with a model you supply

Aider's documentation describes AI pair programming in a local Git repository and lists connections to cloud providers, Ollama, and other compatible routes. Aider's editing, repository-map, and test workflows are client features. They do not make a paid provider free.

Aider is a practical local-first option when you want a small, inspectable change loop. One $0 route is a local model you can already run; a hosted provider may offer its own free tier. Check that provider’s terms separately.

6. Crush: model choice is the feature, not a free quota

The official Crush repository describes a terminal agent with model selection, mid-session switching, LSP context, MCP extensions, and support for OpenAI- or Anthropic-compatible APIs across several platforms. It does not publish a universal provider quota in the client description.

Use Crush when the TUI, session model switching, or platform support matches your workflow. Price the selected API or local model separately.

7. GitHub Copilot CLI: check plan and AI-credit rules

GitHub's Copilot CLI documentation says Copilot CLI is available with all Copilot plans and that each interaction consumes AI credits based on the tokens processed and model used. The current plans page describes Copilot Free as limited access rather than an unlimited CLI allowance.

Copilot CLI also supports BYOK. GitHub's authentication guide says GitHub authentication is not required for BYOK model requests, but cloud delegation, the GitHub MCP server, and GitHub Code Search still require GitHub authentication. That distinction matters when a "free" setup needs GitHub context.

Comparison Table: Access and Cost

Question	Gemini CLI	Codex CLI	OpenCode	goose	aider	Crush	Copilot CLI
Where is the free allowance?	Former consumer allowance moved to Antigravity; do not rely on the legacy quota table	ChatGPT plan, if eligible	Provider, subscription, or local route; no universal client quota	Provider, subscription, or local route	Provider or local route	Provider or local route	Copilot plan and AI-credit allowance
Can the client run against a local project?	Yes, after local authentication	Yes, with local permissions and sandbox settings	Yes	Yes, goose runs on your machine	Yes, local Git repo	Yes, terminal workflow	Yes, from the terminal
Can I use a local model?	The Google access routes discussed here use hosted models	Yes, `--oss` with Ollama or LM Studio; separate from ChatGPT quotas	Provider-dependent	Official site lists Ollama among provider options	Official docs list Ollama	Compatible API or provider-dependent	BYOK or offline mode has separate requirements
What determines cost?	Login method, quota, model family, or Cloud/API billing	ChatGPT plan limits or API token pricing	Provider terms, subscription, or local hardware	Selected provider, subscription, or hardware	Selected provider or hardware	Selected provider or hardware	Plan credits, model usage, or BYOK provider
What should I verify first?	Transition eligibility and continuing account/API access before `/stats model`	`/status`, plan, model, and local/cloud mode	Provider auth route, model availability, and budget or hardware	Provider, recipe, extension, and sandbox settings	Model key, local model, and Git state	Model endpoint, permissions, and platform	Plan, AI credits, GitHub auth, and BYOK settings

Use this table to identify the account, provider, or local service you need before installing.

How to Read the Comparison

A client, model route, account plan, and hardware setup are different variables. Results from different benchmark variants or configurations cannot rank every other setup. Compare the conditions that match your work:

Gemini CLI requires a continuing supported enterprise or paid API route; personal users should evaluate Antigravity CLI instead.
Codex CLI is the ChatGPT or API-key route when its plan and local permission model fit your work.
OpenCode, goose, aider, and Crush are free client software; choose a provider, subscription, or local model route and check its cost and limits.
Copilot CLI is the GitHub-integrated route, bounded by the current Copilot plan and credit rules.

Choose by Scenario

Scenario	Start with	Why	Cost checkpoint
You want Google's individual-account route	Antigravity CLI	It replaces Gemini CLI for the affected consumer accounts	Check current plan eligibility and limits; do not reuse Gemini CLI's old daily quota
You already use ChatGPT for coding	Codex CLI	The local CLI, plan access, API-key path, and cloud handoff are documented separately	Free is limited; subscription limits and API charges are not the same
You want to switch providers	OpenCode	The client is provider-configurable	Check whether the provider offers OAuth/subscription access, API billing, or a local endpoint; set the relevant budget or hardware boundary
You connect tools and repeatable workflows	goose	Extensions, recipes, MCP, and subagents are first-class documented concepts	Provider and extension permissions still apply
You want focused edits in Git	aider	It is designed around pair programming and local Git repositories	Model cost is separate from the client
You want a model-switchable TUI	Crush	LSP, MCP, sessions, and compatible APIs are documented features	The selected endpoint determines the bill
Your work is GitHub-first	Copilot CLI	CLI, GitHub context, plan mode, and GitHub integrations are documented	Copilot plan credits and GitHub authentication determine access
You need an offline or private fallback	An open client plus a local model	Avoids a cloud API route when your hardware can run the model	Hardware, electricity, storage, and local model quality are real costs

No row is a universal recommendation. It is a starting point that matches an access condition.

Practical $0 Setups and Their Boundaries

Account-quota setup

Use limited Codex or Copilot access when your account is eligible. For Google, check Antigravity's current individual plan rather than Gemini CLI's old free-account route. Keep each provider's quota separate, and do not add legacy request counts to current ChatGPT access or Copilot credits as one daily total.

Local-model setup

Pair a local model service such as Ollama with OpenCode, goose, aider, or Crush. The client may be free and there may be no per-request cloud bill, but the machine must still have enough memory, storage, and power. Verify the model's license and performance for your task before treating it as a production fallback.

Escalation setup

Use an account quota for exploration and reserve a paid API or subscription for work that needs a different model or higher limits. Monitor provider usage and keep the fallback explicit. The AI CLI cost optimization guide can be used after you have measured the route you actually use.

If you run more than one client, Termdock can keep terminal panes visible side by side while you switch projects and review the correct working tree.

Check the Official Setup Guide

Before installing, open the provider's current installation and authentication page. The command and package can change independently of the quota:

After authentication, check the provider's own status command where available. Antigravity's quickstart documents /usage; continuing Gemini CLI users can use /stats model; Codex documents /status. Do not paste API keys into a project file or assume a successful client install means model access is free.

Compare Your Actual Usage in Termdock

Choose two CLIs you have confirmed you can access. In separate Termdock terminals, test the same low-risk task against the same acceptance criteria. Record each account path, model, output, actual quota or API usage, and manual corrections. Sessions do not share prompts; give each the same inputs explicitly and do not paste API keys into terminal output. Check each provider's actual bill before deciding whether an option is free or less expensive for you.

When Free Is Not Enough

Free access is no longer the right constraint when:

the daily or monthly limit interrupts a repeatable workflow;
local hardware cannot run the model with acceptable context or latency;
provider terms do not fit the repository's data policy;
a team needs centralized billing, audit, or access controls; or
you spend more time changing providers than reviewing the code.

At that point, compare a paid subscription, a pay-as-you-go API, and a managed enterprise route using current provider pages. The decision is about the required model, controls, and workflow, not about preserving a "free" label.

Sources Checked on 2026-10-27

Danny Huang·Follow on Threads →

#ai-cli#gemini-cli#codex-cli#opencode#goose#aider#free-tools

Free AI CLI Tools in 2026: Quotas, local models, and how to choose