FreeLLMAI apps: free and paid New
The apps people use with AI models: coding agents, code editors, apps to run models at home, to fine-tune them, and to build apps. For each one: what you can use free, whether it works with free or local models, and, where tests exist, which gives the best results and uses the fewest tokens.
- Free The app is free; no plan to buy.
- Free with your model The app is free; you plug in a model: a free one, a local one, or your own paid key.
- Free plan with limits There is a free plan, with a monthly or daily limit; more needs a paid plan.
- Paid No free plan.
Coding agents in the terminal
| App | Price | What is free | Open source | Free or local models |
|---|---|---|---|---|
| OpenCode | Free with your model | Free app. Its Zen gateway lists free models “for a limited time”. | MIT | Yes |
| Codex CLI | Free plan with limits | Works with ChatGPT Free (limits not published), ChatGPT Go from $8, or your own API key. | Apache-2.0 | Yes |
| Aider | Free with your model | Free app; you pay only the model, or use a local one. | Apache-2.0 | Yes |
| Goose | Free with your model | Free app; desktop and terminal. | Apache-2.0 | Yes |
| Qwen Code | Free with your model | Its free sign-in tier ended on 15 Apr 2026: now your own key or a local model. | Apache-2.0 | Yes |
| Crush | Free with your model | Free app; its Hyper provider has a free tier (limits not published). | FSL-1.1-MIT (source-available) | Yes |
| Kilo Code | Free with your model | Free plan with your own key and an “Auto Free” mode that picks free models; Kilo Pass $19 a month. | MIT | Yes |
| Cline | Free with your model | Free app for VS Code, JetBrains and the terminal. | Apache-2.0 | Yes |
| Kimi Code | Price not verified | Open source; pricing not verified. (partly unverified) | MIT | No |
| Gemini CLI | Free with your model | Free Google sign-in use ended on 18 Jun 2026; Google points to Antigravity CLI instead. Whether a free API key still works is unclear. (partly unverified) | Apache-2.0 | No |
| Antigravity CLI | Free plan with limits | $0 plan with weekly limits (numbers not published). (partly unverified) | closed | No |
| GitHub Copilot CLI | Free plan with limits | Included in Copilot Free; Copilot Pro $10 a month. | closed | No |
| Amp | Free plan with limits | Free Hobby plan with pay-as-you-go; $20 plan. (partly unverified) | closed | No |
| Cursor CLI | Free plan with limits | Free tier “limited” (no numbers); Plus $8 a month. | closed | No |
| Claude Code | Paid | From Claude Pro, $20 a month ($17 billed yearly). Works only with Claude models. | closed | No |
AI code editors
| App | Price | What is free | Open source | Free or local models |
|---|---|---|---|---|
| Zed | Free with your model | Personal plan $0, unlimited with your own key or local models; Pro $10 a month. | GPL-3.0 / Apache-2.0 | Yes |
| VS Code + GitHub Copilot | Free plan with limits | Copilot Free: 2,000 completions and a limited number of chats a month; Pro $10. Chat and agent also work with Ollama or your own key. | VS Code MIT, Copilot closed | Yes |
| Trae | Free plan with limits | Free: 5,000 autocompletions a month and a limited Auto mode; Pro $20. | closed | Yes |
| Cursor | Free plan with limits | Free: “limited” requests (no numbers); Plus $8, Premium $30. Own keys only for chat. | closed | No |
| JetBrains AI (Junie) | Free plan with limits | AI Free: 3 credits every 30 days; works with your own key, Ollama or LM Studio. | closed | Yes |
| Kiro | Free plan with limits | Free: 50 credits a month; Pro $20 (1,000 credits). Managed models only. | closed | No |
| Google Antigravity | Free plan with limits | $0 with weekly limits (numbers not published); Gemini-centric. (partly unverified) | closed | No |
| Devin Desktop (ex Windsurf) | Free plan with limits | Free “light” quota (numbers not published); Pro $20. Renamed from Windsurf on 2 Jun 2026. (partly unverified) | closed | No |
Run models at home
| App | Price | What is free | Open source | Free or local models |
|---|---|---|---|---|
| LM Studio | Free | Free for personal and internal business use; paid plans add cloud models. Easiest app to start with. | closed | Yes |
| Ollama | Free | Free on your own computer; optional cloud plans (Pro $20). | MIT | Yes |
| Jan | Free | Free desktop app, works offline. | Apache-2.0 | Yes |
| llama.cpp | Free | The engine most other apps use; for technical users. | MIT | Yes |
| AnythingLLM | Free | Free; chat with your own documents. | MIT | Yes |
| Open WebUI | Free | Free web interface for Ollama and others; branding rules above 50 users. | Custom license (source-available) | Yes |
| KoboldCpp | Free | Free; popular for stories and role-play. | AGPL-3.0 | Yes |
| LocalAI | Free | Free; an OpenAI-compatible server on your machine. | MIT | Yes |
| vLLM | Free | Free; fast serving for servers and many users. | Apache-2.0 | Yes |
| Lemonade | Free | Free; optimised for AMD hardware. | Apache-2.0 | Yes |
| Msty Studio | Free plan with limits | Free with feature limits; Aurum $149 a year. (partly unverified) | closed | Yes |
| GPT4All | Free | Free, but the last release is from Feb 2025. | MIT | Yes |
Fine-tune models at home
| App | Price | What is free | Open source | Free or local models |
|---|---|---|---|---|
| Unsloth Studio | Free | Free no-code web app (beta) to train, run and export models (GGUF, MLX) on your computer. | AGPL-3.0 (core Apache-2.0) | Yes |
| Unsloth | Free | Free library: faster fine-tuning with less memory. | Apache-2.0 | Yes |
| LLaMA-Factory | Free | Free, with a web interface. | Apache-2.0 | Yes |
| Axolotl | Free | Free; Linux with NVIDIA or AMD. | Apache-2.0 | Yes |
| TRL | Free | Free Hugging Face library; for developers. | Apache-2.0 | Yes |
App and design builders
| App | Price | What is free | Open source | Free or local models |
|---|---|---|---|---|
| Open Design | Free | Free; drives the coding agents you already have. | Apache-2.0 | Yes |
| bolt.diy | Free with your model | Free, open version of Bolt; your key or a local model. | MIT | Yes |
| Dyad | Free with your model | Free with your own key; Pro $20 a month. | Apache-2.0 (except Pro parts) | Yes |
| Onlook | Price not verified | Open source, can be self-hosted; hosted price not verified. (partly unverified) | Apache-2.0 | No |
| Bolt.new | Free plan with limits | Free: 1M tokens a month (300K a day); Pro from $18. | closed | No |
| v0 | Free plan with limits | Free: 7 messages a day plus a one-time $5 credit; Plus $30. | closed | No |
| Lovable | Free plan with limits | Free: 5 credits a day, up to 30 a month; Pro $25. | closed | No |
| Replit Agent | Free plan with limits | Free daily credits (numbers not published); Core $20. | closed | No |
| Google AI Studio Build | Free plan with limits | Free quota (limits not published); Gemini models only. (partly unverified) | closed | No |
Which coding agents give the best results
Terminal-Bench 2.0: tasks solved (%). Higher is better.
What this shows · How to read it
What it shows. The share of real terminal tasks each coding agent solved, with the model it was tested with.
How to read it. Longer bar, more tasks solved. The same app can score very differently with another model, so read it as app plus model. Click a bar to see the app’s price and free plan above.
Real terminal tasks, success rate. Each row is an app with a specific model, so it scores the pair, not the app alone. Submissions closed in May 2026, so newer models are missing. With the same model (Opus 4.6), the app changes the score by up to 17 points: Terminus-KIRA 74.7%, Terminus 2 62.9%, Claude Code 58.0%. Source: Terminal-Bench 2.0.
Which use the fewest tokens
Million tokens used per coding task.
What this shows · How to read it
What it shows. How many tokens each coding agent used, on average, to do one coding task, and what that costs at API prices.
How to read it. Shorter bar, fewer tokens. Fewer tokens means cheaper and usually faster, but check the score in the table: an agent that uses few tokens and fails the task is not a saving.
| App | Model | Score | Tokens per task | Cost per task |
|---|---|---|---|---|
| Claude Code | Sonnet 5.5 (max) | 68.4 | 27.7M | $14.19 |
| Devin Fusion CLI | — | 68.1 | 12.8M | $9.01 |
| Claude Code | Opus 5.5 (max) | 66 | 15.6M | $13.04 |
| Antigravity CLI | Gemini 4 Argon | 63.8 | 13.7M | $5.84 |
| Claude Code | Sonnet 5.5 (xhigh) | 62.9 | 7.1M | $3.33 |
| Codex | GPT-6.1 Sol (xhigh) | 62.9 | 3.2M | $1.04 |
| Codex | GPT-6.1 Sol (medium) | 61.4 | 2.3M | $0.70 |
| Grok Build | Grok 4.7 | 56.3 | 14.3M | $8.82 |
| OpenCode | GLM-5.3 | 53.6 | 14.3M | $4.24 |
| Kimi Code CLI | Kimi K3 | 51.9 | 8.5M | $5.05 |
| Claude Code | Sonnet 5.5 (medium) | 45.9 | 1.3M | $0.62 |
| Codex | DeepSeek V4 Flash | 38.7 | 15.4M | $0.09 |
Coding tasks from three benchmarks. Tokens and dollars per task at pay-per-token API prices; on a subscription your real cost is different. Effort level matters a lot: the same app can use 20 times more tokens at maximum effort. Score: the Artificial Analysis Coding Agent Index (higher is better). Source: Artificial Analysis Coding Agent Index. Apart from this index, we found no rigorous public test of token use app by app.
No longer maintained
- Roo Code: closed on 15 May 2026, repository archived.
- Continue: no longer maintained.
- Void: archived on 2 Jun 2026.
- Kimi CLI: archived; replaced by Kimi Code.
Prices and free limits checked on the official pages on 20 Oct 2026. Free plans change often: check the app’s own page before you rely on a limit. We are not paid by any of these apps.