What free AI coding tools exist in 2026? One table for every quota
Most developers assume free AI coding starts and ends with GitHub Copilot's 2,000 completions. In reality, June 2026 stacks three resource pools—overseas CLIs, open-source aggregators, and domestic API platforms—that together cover daily coding, refactors, and automation scripts. The table below reflects official public quotas as of 2026-06-09 (policies change; verify on each vendor site).
| Tool / platform | Free quota | Card required | VPN-free | Best leverage |
|---|---|---|---|---|
| Gemini CLI (OAuth) | 1,000/day, 60 RPM | No | VPN needed | No API key; usable until June 18 |
| Antigravity CLI | Free public beta (individual) | No | VPN needed | Official Gemini CLI successor |
| Codex CLI | ChatGPT free account (limited) | No | No | Route domestic API, zero subscription |
| OpenCode | Free tool + BYOK | No | Yes | 75+ providers, 146K GitHub stars |
| GitHub Copilot Free | 2,000 completions + 50 premium/mo | No | Yes | Enable from GitHub Settings |
| Cursor Hobby | 2,000 Tab + 50 slow/mo | No | Yes | Full VS Code fork experience |
| SiliconFlow | 20M tokens (permanent) | No | Yes | DeepSeek, Qwen, GLM-5 |
| Alibaba Bailian | 70M tokens (time-limited) | No | Yes | 70+ models, full Qwen 3.5 line |
| Zhipu AI | 20M tokens (permanent) | No | Yes | GLM-5, Claude-compatible routing |
Hidden rate-limit traps: Free tiers cap RPM and RPD. Parallel agents or whole-repo scans burn quota in minutes and trigger 429 throttling.
OAuth vs API key pools: Gemini CLI OAuth quotas and AI Studio API key quotas are counted separately. You can run both in parallel, but routing OAuth tokens through third-party proxies violates Google's ToS.
June 18 migration deadline: Google ends individual free-tier Gemini CLI service on June 18, 2026. Move workflows to Antigravity CLI; Standard and Enterprise licenses stay on Gemini CLI. Read our shutdown analysis for context.
VPN-free path: SiliconFlow plus OpenCode or Codex CLI is the most stable zero-cost stack for developers who cannot rely on overseas network access. Endpoints are OpenAI-compatible—two environment variables and you are live.
Student multiplier: GitHub Education unlocks Copilot Pro-equivalent access (roughly $10/month value). Cursor offers one free Pro year for verified students. Copilot new-signup pauses have appeared in 2026; check the live button on GitHub Settings.
Treat the table as a resource map, not a single-vendor bet. Overseas OAuth suits experimentation and English-first projects. Domestic APIs suit long-running stable calls. IDE free tiers (Cursor, Copilot) cover completions and light chat. The real win is routing task types to the right tool instead of forcing everything through one paid subscription.
Gemini CLI, Codex CLI, Claude Code: how to run overseas terminal agents for free
Terminal agents dominate developer discourse in 2026. Each tool has a distinct free path, install command, and model-switch workflow.
Gemini CLI / Antigravity CLI: Still the most generous official CLI free tier. Run npm install -g @google/gemini-cli, launch gemini, choose Sign in with Google, and you receive 1,000 daily requests at 60 RPM with Gemini 2.5 Pro/Flash and a 1M-token context window. Use /model inside a session to switch models and /stats model to track daily burn. After June 18, individual users should install Antigravity CLI (agy) and migrate—public beta remains free for personal use and retains Agent Skills, Hooks, and Subagents.
Codex CLI: OpenAI's Rust terminal agent with 83K+ GitHub stars and kernel-level sandboxing. The free path is ChatGPT OAuth (codex then Sign in with ChatGPT). For developers who need a more practical route, point Codex at a domestic API in ~/.codex/config.toml:
openai_base_url = "https://api.siliconflow.cn/v1" model = "deepseek-ai/DeepSeek-V3" sandbox_mode = "workspace-write" web_search = "disabled" approval_policy = "on-request"
Set export OPENAI_API_KEY="sk-xxxx", run codex doctor, and you have a zero-subscription agent on domestic infrastructure. One-shot tasks work via codex "fix this build error" in non-interactive mode.
Claude Code: Code quality benchmark, but the official free path requires Pro or Max ($20+/month). The budget alternative is a domestic relay API in ~/.claude/settings.json—Zhipu GLM-5 or Claude-compatible models on SiliconFlow both work. Never run /init on a full repository; that single command can consume tens of thousands of tokens.
Overseas CLI free tiers share the same weakness: network dependency, OAuth refresh failures after laptop sleep, and no true 24/7 uptime for long agent jobs—the gap a cloud Mac host fills later in this guide.
OpenCode, OpenClaw, Cursor, Copilot: open aggregators vs IDE free tiers
OpenCode (146K stars) is free and open source. Connect providers through /connect or ~/.config/opencode/config.json—Gemini API Flash (1,500/day), SiliconFlow DeepSeek-V3, or local Ollama. Inside the TUI, /models switches models and Tab toggles Plan/Build mode. It is one of the most flexible zero-subscription multi-model entry points available.
OpenClaw reuses Gemini CLI OAuth (openclaw models auth login --provider google-gemini-cli) and Claude setup-token flows. It targets Gateway hosts that stay online and bridge Telegram, Slack, or custom channels—closer to an automation shell than OpenCode's pure terminal UX.
| IDE free tier | Completion quota | Premium requests | Best for |
|---|---|---|---|
| Cursor Hobby | 2,000 Tab/month | 50 slow/month | VS Code users evaluating agents |
| GitHub Copilot Free | 2,000/month | 50 premium/month | Native GitHub PR and issue flow |
| Copilot Student | Unlimited completions | 300 premium/month | Verified .edu accounts |
Download Cursor from cursor.com; Hobby needs no credit card. Track usage at app.cursor.sh/account/usage. Enable Copilot Free under GitHub Settings → Copilot. Students can apply for one free Pro year at cursor.com/students. If you already use Cursor Agent Skills, our complete Skills guide helps you extract more from the free tier.
Domestic API quick reference: SiliconFlow https://api.siliconflow.cn/v1; Alibaba Bailian https://dashscope.aliyuncs.com/compatible-mode/v1; Zhipu https://open.bigmodel.cn/api/paas/v4. All three accept phone or email signup with no card binding.
From signup to first agent: six-step free AI coding runbook
Audit needs and network: Stable overseas access → prioritize Gemini CLI or Antigravity CLI OAuth. VPN-free only → register SiliconFlow or Bailian and grab an API key first.
Install a terminal agent: Domestic path: curl -fsSL https://opencode.ai/install | bash or npm i -g @openai/codex. Overseas path: npm i -g @google/gemini-cli plus Google OAuth.
Wire API endpoints: Set OPENAI_API_KEY and OPENAI_BASE_URL (or tool-specific config files). Validate with codex doctor or OpenCode /connect.
Pick default models wisely: Flash or Lite for routine tasks; Pro or DeepSeek-V3 for hard refactors. Defaulting to the strongest model drains quota fastest.
Layer IDE free tiers: Let Copilot or Cursor handle Tab completion while the terminal agent handles cross-file refactors—quotas do not overlap.
Set alerts and migration plan: Enable 80% quota email alerts in cloud consoles. Before June 18, migrate Gemini CLI to Antigravity CLI and back up ~/.gemini configs.
Three hard numbers plus ten token-saving tactics to stretch free tiers three months
1,000/day × 30 = 30,000/month: Gemini CLI OAuth at this rate exceeds the practical daily ceiling of a $20 Pro subscription—but only for personal terminal use; proxy forwarding is prohibited.
90M tokens across three domestic platforms: SiliconFlow 20M + Bailian 70M + Zhipu 20M (some time-limited) can sustain months of DeepSeek-V3 API calls on a mid-size project at current free-tier pricing.
2,000 + 2,000 dual IDE completions: Cursor Hobby and Copilot Free each grant 2,000 Tab completions monthly—4,000 combined, enough for a personal side project end to end.
Token discipline boils down to a few rules. Ban whole-repo scans (Claude /init, Codex wide reads). Prefer single-file edits over "optimize the entire codebase." Run Flash first—Gemini 2.5 Flash-Lite allows 1,500 daily calls for triage before escalating to Pro. Offload formatting and comment generation to local Ollama with qwen3:8b. Keep system prompts short. Watch vendor campaigns for bonus vouchers during launch weeks and anniversaries.
Note: Free policies shift without notice. Groq (14,400/day) and NVIDIA NIM free tiers require VPN access. Data captured on 2026-06-09; confirm current terms on each vendor site.
Free tiers excel at learning, prototyping, and side projects. They break down when you need 24/7 agent uptime, parallel sub-agents, Xcode CI, or iOS builds—laptop sleep kills OAuth sessions, memory swap slows responses, and cheap Linux VPS hosts lack native Apple Silicon support for Xcode and Seatbelt sandboxes. Teams that need production-grade AI coding environments, OpenClaw Gateway persistence, or multi-region collaboration usually land on MESHLAUNCH Mac Mini cloud rental: dedicated Apple Silicon, day/week/month billing, and a clean pairing with zero-cost API layers. For host sizing against CLI workloads, see our OpenRouter CLI rankings and Mac rental guide.
Yes—OAuth login grants 1,000 requests daily at 60 RPM with no card. From June 18, individual free tiers move to Antigravity CLI; install and sign in early. Enterprise license holders keep Gemini CLI. Cloud Mac deployment details are on our pricing page.
Register at SiliconFlow, Alibaba Bailian, or Zhipu AI for a free API key, then configure OpenCode or Codex CLI with the domestic endpoint. Cursor Hobby and Copilot Free work directly without proxy setup. Network and deployment questions: help center.
Quotas are nearly identical (2,000 completions plus 50 premium requests monthly). Cursor wins on Agent and Composer UX; Copilot wins on native GitHub PR and issue integration. Students should enable both and apply for GitHub Education to unlock Copilot Pro and Cursor Pro free year.
Yes, in parallel on different tools—Gemini CLI for OAuth-backed tasks, OpenCode or Codex CLI for SiliconFlow-backed tasks. Quotas are independent. Do not proxy Gemini OAuth tokens to third parties; that violates Google's terms. For migration timing, read our Gemini CLI shutdown guide.