Free and open-source alternatives to AI subscriptions
Open models and apps you can run yourself, sorted by the setup you have, plus hosted services if you don’t want to run anything. Licences, hardware needs and hosted prices are from each project’s or host’s own pages, checked .
Chat · Search and research · Coding · Writing · Images · Design · Video · Voice (text to speech) · Music · Transcription · Upscaling · Productivity · All-in-one creative tools · Personal ai agents
Free chat options
Not ranked: we haven’t recorded a public leaderboard for this use yet.
| Your setup | Option | Licence or price | Needs / free tier |
|---|---|---|---|
| Windows or Linux PC with an NVIDIA GPU or Apple Silicon Mac | Qwen3 (8B) ↗ | Apache-2.0 | 8B ≈ 6–8GB VRAM at 4-bit; runs on 16GB Macs |
| CPU only or a laptop | Ollama ↗ | MIT | Runs on macOS/Windows/Linux; RAM needs scale with model (8B ≈ 8–16GB RAM) |
| No hardware (hosted) | OpenRouter (free models) ↗ | Pay per token for paid models; ':free' model variants $0 | Yes: free models 20 req/min, 50/day (1000/day after $10 lifetime credit purchase) |
Free search and research options
Not ranked: we haven’t recorded a public leaderboard for this use yet.
| Your setup | Option | Licence or price | Needs / free tier |
|---|---|---|---|
| No hardware (hosted) | OpenRouter (free models) ↗ | Pay per token for paid models; ':free' model variants $0 | Yes: free models 20 req/min, 50/day (1000/day after $10 lifetime credit purchase) |
Free coding options
Not ranked: we haven’t recorded a public leaderboard for this use yet.
| Your setup | Option | Licence or price | Needs / free tier |
|---|---|---|---|
| Windows or Linux PC with an NVIDIA GPU or Apple Silicon Mac | Qwen3-Coder 30B-A3B ↗ | Apache-2.0 | MoE 3B active; ~20GB at 4-bit, 32GB Mac |
| CPU only or a laptop | Ollama ↗ | MIT | Runs on macOS/Windows/Linux; RAM needs scale with model (8B ≈ 8–16GB RAM) |
| No hardware (hosted) | OpenRouter (free models) ↗ | Pay per token for paid models; ':free' model variants $0 | Yes: free models 20 req/min, 50/day (1000/day after $10 lifetime credit purchase) |
Free writing options
Not ranked: we haven’t recorded a public leaderboard for this use yet.
| Your setup | Option | Licence or price | Needs / free tier |
|---|---|---|---|
| Windows or Linux PC with an NVIDIA GPU or Apple Silicon Mac | Qwen3 (8B) ↗ | Apache-2.0 | 8B ≈ 6–8GB VRAM at 4-bit; runs on 16GB Macs |
| CPU only or a laptop | LanguageTool (self-host) ↗ | LGPL-2.1 | Java; 2–4GB RAM server |
| No hardware (hosted) | OpenRouter (free models) ↗ | Pay per token for paid models; ':free' model variants $0 | Yes: free models 20 req/min, 50/day (1000/day after $10 lifetime credit purchase) |
Free images options
Not ranked: we haven’t recorded a public leaderboard for this use yet.
| Your setup | Option | Licence or price | Needs / free tier |
|---|---|---|---|
| Windows or Linux PC with an NVIDIA GPU, Apple Silicon Mac or CPU only or a laptop | ComfyUI ↗ | GPL-3.0 | As low as 4GB VRAM + 8GB RAM (offloading); NVIDIA/AMD/Intel/Apple Silicon; desktop app Win/macOS |
| No hardware (hosted) | Replicate ↗ | Pay per use. FLUX dev $0.025/img, FLUX schnell $3/1000 imgs, Wan 2.1 i2v 480p $0.09/s video; T4 $0.81/hr, L40S $3.51/hr, H100 $5.49/hr | No standing free tier stated on pricing page |
Free design options
Not ranked: we haven’t recorded a public leaderboard for this use yet.
| Your setup | Option | Licence or price | Needs / free tier |
|---|---|---|---|
| Windows or Linux PC with an NVIDIA GPU, Apple Silicon Mac or CPU only or a laptop | ComfyUI ↗ | GPL-3.0 | As low as 4GB VRAM + 8GB RAM (offloading); NVIDIA/AMD/Intel/Apple Silicon; desktop app Win/macOS |
| No hardware (hosted) | fal.ai ↗ | Pay per use. Z-Image Turbo $0.005/MP, ACE-Step music $0.0002/s; GPUs H100 from $2.49/hr | Not stated on pricing page |
Free video options
Not ranked: we haven’t recorded a public leaderboard for this use yet.
| Your setup | Option | Licence or price | Needs / free tier |
|---|---|---|---|
| Windows or Linux PC with an NVIDIA GPU or CPU only or a laptop | ComfyUI ↗ | GPL-3.0 | As low as 4GB VRAM + 8GB RAM (offloading); NVIDIA/AMD/Intel/Apple Silicon; desktop app Win/macOS |
| Apple Silicon Mac | LTX-Video ↗ | LTXV Open Weights License (custom) | Distilled/fp8 variants low VRAM; macOS MPS supported |
| No hardware (hosted) | RunPod ↗ | Pods: RTX 3090 $0.22 (community)–$0.50/hr, RTX 4090 $0.34–$0.89/hr, A100 80GB $1.19–$1.79/hr; Serverless 24GB $0.69/hr; public endpoints e.g. FLUX schnell $0.0024/MP, Wan 2.2 720p 5s $0.30 | No |
Free voice (text to speech) options
Top open pick: Fish Audio Drama 3, the highest-ranked open-weights model on the Artificial Analysis leaderboard ↗ (). Drama 3 is still labelled Preview on AA and is too big for most home PCs; Kokoro is the run-it-yourself pick for CPU/Mac/low-VRAM. Chatterbox (MIT) is the fully permissive pick.
| Your setup | Option | Licence or price | Needs / free tier |
|---|---|---|---|
| Windows or Linux PC with an NVIDIA GPU | Fish Audio Drama 3 (Fish Speech repo) ↗ | Open weights; free commercial use under US$500K yearly revenue; licence needed to offer it as a TTS/cloning service; no cloning real people without consent | Large: 35B MoE (3B active). Needs a big GPU; most people should use a hosted option |
| Apple Silicon Mac or CPU only or a laptop | Kokoro-82M ↗ | Apache-2.0 | Tiny (82M); runs on CPU; Apple Silicon MPS |
| No hardware (hosted) | Together AI ↗ | Per token/unit. gpt-oss-120B $0.15/$0.60 per 1M tok; Llama 3.3 70B $1.04; Kokoro-82M TTS $4/1M chars; Whisper Large v3 $0.0015/min; FLUX.2 dev $0.0154/img | Some $0 models listed (e.g. Ternary Bonsai 27B) |
Free music options
Not ranked: we haven’t recorded a public leaderboard for this use yet.
| Your setup | Option | Licence or price | Needs / free tier |
|---|---|---|---|
| Windows or Linux PC with an NVIDIA GPU | YuE ↗ | Apache-2.0 | Linux + NVIDIA 24GB VRAM (BF16); no Mac |
| Apple Silicon Mac | ACE-Step ↗ | Apache-2.0 | Max VRAM reduced to 8GB; GUI on Win/macOS/Linux |
| No hardware (hosted) | fal.ai ↗ | Pay per use. Z-Image Turbo $0.005/MP, ACE-Step music $0.0002/s; GPUs H100 from $2.49/hr | Not stated on pricing page |
Free transcription options
Not ranked: we haven’t recorded a public leaderboard for this use yet.
| Your setup | Option | Licence or price | Needs / free tier |
|---|---|---|---|
| Windows or Linux PC with an NVIDIA GPU or Apple Silicon Mac | Whisper (OpenAI) ↗ | MIT | tiny ~1GB to large ~10GB VRAM; macOS ok |
| CPU only or a laptop | whisper.cpp ↗ | MIT | CPU, Apple Silicon optimized |
| No hardware (hosted) | Together AI ↗ | Per token/unit. gpt-oss-120B $0.15/$0.60 per 1M tok; Llama 3.3 70B $1.04; Kokoro-82M TTS $4/1M chars; Whisper Large v3 $0.0015/min; FLUX.2 dev $0.0154/img | Some $0 models listed (e.g. Ternary Bonsai 27B) |
Free upscaling options
Not ranked: we haven’t recorded a public leaderboard for this use yet.
| Your setup | Option | Licence or price | Needs / free tier |
|---|---|---|---|
| Windows or Linux PC with an NVIDIA GPU or Apple Silicon Mac | Upscayl ↗ | AGPL-3.0 | Vulkan GPU required; macOS 12+, Win, Linux |
| No hardware (hosted) | Replicate ↗ | Pay per use. FLUX dev $0.025/img, FLUX schnell $3/1000 imgs, Wan 2.1 i2v 480p $0.09/s video; T4 $0.81/hr, L40S $3.51/hr, H100 $5.49/hr | No standing free tier stated on pricing page |
Free productivity options
Not ranked: we haven’t recorded a public leaderboard for this use yet.
| Your setup | Option | Licence or price | Needs / free tier |
|---|---|---|---|
| Windows or Linux PC with an NVIDIA GPU | LM Studio ↗ | Proprietary app, free (CLI 'lms' MIT) | Apple Silicon M1+ macOS 14+, 16GB+ RAM rec.; Windows AVX2, 16GB RAM, 4GB+ VRAM rec. |
| Apple Silicon Mac or CPU only or a laptop | Ollama ↗ | MIT | Runs on macOS/Windows/Linux; RAM needs scale with model (8B ≈ 8–16GB RAM) |
| No hardware (hosted) | OpenRouter (free models) ↗ | Pay per token for paid models; ':free' model variants $0 | Yes: free models 20 req/min, 50/day (1000/day after $10 lifetime credit purchase) |
Free all-in-one creative tools options
Not ranked: we haven’t recorded a public leaderboard for this use yet.
| Your setup | Option | Licence or price | Needs / free tier |
|---|---|---|---|
| Apple Silicon Mac | Jan ↗ | Apache-2.0 (GitHub reports NOASSERTION) | Desktop app, Mac/Win/Linux |
| No hardware (hosted) | Poe ↗ | Subscriptions from $4.99/mo; add-on points ~$30 per 1M | Yes: 300 compute points/day in most markets since 2026-03-30 (from Poe email quoted on Reddit; official FAQ confirms daily points, not amount) |
Free personal AI agents options
Not ranked: we haven’t recorded a public leaderboard for this use yet.
| Your setup | Option | Licence or price | Needs / free tier |
|---|---|---|---|
| Apple Silicon Mac or CPU only or a laptop | Hermes Agent (Nous Research) ↗ | MIT | Runs on macOS/Windows/Linux CPU; pair with a local model (RAM needs scale with model) or a hosted API |
Hosted prices are in USD. A licence marked non-commercial means you can’t use the output for paid work. Hardware figures for some models are typical 4-bit quantized sizes; check the model page before you buy hardware.