No rows match that filter.
CHT
Chat assistants
8 tracked| Service | What's free | Limits | The catch | Verified |
|---|---|---|---|---|
| ChatGPTchatgpt.comno card needed | Flagship auto model, file uploads, image generation, a small daily budget of thinking-mode messages. | ~10 msgs / 5 h fast window, then auto-downgrade to mini model | Limits are dynamic. Opting out of personalized ads can cut the cap to as low as ~5 msgs / 3 h. | Jul 2 |
| Claudeclaude.aino card needed | Claude Sonnet (current gen) on web, iOS, Android and desktop. File uploads and artifacts included. | ~15-40 msgs per 5 h session window, varies with demand | Caps shrink at peak hours. Long conversations burn the quota much faster. | Jul 4 |
| DeepSeekchat.deepseek.comno card needed | Full flagship model including the reasoning mode, on web and mobile. | No published caps | Data is processed in China and chats are used for training by default. | Jul 2 |
| Perplexityperplexity.ai | Unlimited quick AI searches with citations. | ~5 Pro searches / day, a few Deep Research runs / day | No Labs or premium model picker on the free plan; Deep Research is capped to a handful of runs a day. | Jul 4 |
| Geminigemini.google.com | Gemini Flash chat, limited Pro access, image generation, 5 Deep Research uses/month, Google app integrations. | Compute-based caps: refresh every 5 h, plus a weekly ceiling | Limits scale with prompt complexity and chat length, so heavy use runs out sooner. Top models and higher caps need a Google AI plan. | Jul 2 |
| Microsoft Copilotcopilot.microsoft.com | GPT-class chat with web grounding, voice, and image generation (15 fast boosts/week, slow queue after). | Dynamic throttling at peak times; images 1024x1024 only | Free image output is licensed for personal, non-commercial use. | Jul 2 |
| Le Chat (Mistral)chat.mistral.ai | Top Mistral models, web search, image generation, code interpreter, 40+ connectors. | ~25 messages/day soft cap | No Mistral Large, Flash Answers or no-telemetry mode on the free plan. | Jul 2 |
| Grokgrok.com / x.com | Grok text chat on grok.com and X for logged-in users. | ~10 messages / 2 h | Needs an X account 7+ days old with a verified phone number. Image generation is paid-only. | Jul 2 |
API
LLM APIs for developers
9 trackedPermanent free tiers only. One-time trial credits are flagged in the catch column. Stack several providers and you can prototype all month for $0.
| Service | What's free | Limits | The catch | Verified |
|---|---|---|---|---|
| Gemini APIai.google.devno card needed | Real ongoing free tier on Gemini Flash models via AI Studio, including multimodal. | Flash only, e.g. Gemini 3 Flash ~1,500 req/day; exact per-project quotas shown in AI Studio | Pro models were pulled from the free tier in April 2026. Enabling billing on the project kills the free tier instantly. Free-tier data may be used for training. | Jul 4 |
| Groqgroq.comno card needed | OpenAI-compatible API. Llama 3.3 70B and others at ~300 tokens/sec. | ~30 req/min, 1,000 req/day, 100K tokens/day (llama-3.3-70b) | Free limits were cut hard in 2026 (daily requests dropped from 14,400 to 1,000 on most models). | Jul 2 |
| Cerebrascerebras.aino card needed | Fastest inference around (~2,000 tokens/sec) on a rotating lineup of open-weight models (Llama, Qwen, GPT-OSS). | ~1M tokens/day | Small model lineup that rotates often: a model you rely on can disappear from the free tier. | Jul 4 |
| Mistral APIconsole.mistral.ai | Experiment tier: roughly 1B tokens/month across Mistral models. One of the most generous quotas anywhere. | ~1B tokens/month, ~50K tokens/min | You must opt in to letting Mistral train on your data to get the free tier. | Jul 2 |
| OpenRouteropenrouter.ai | Dozens of ":free" models from many labs behind one OpenAI-compatible key. | 50 req/day (20 req/min) across free models | A one-time $10 credit purchase lifts the cap to 1,000 req/day. Free models may log or train on prompts and can vanish without notice. | Jul 4 |
| GitHub Modelsgithub.com/marketplace/models | Playground and API access to GPT, Llama and more with any GitHub account. | High-tier models 50 req/day (10/min), mini models 150 req/day; 8K tokens in / 4K out per request | Prototyping only: per-request token caps are tight and concurrency is 2. | Jul 2 |
| Coherecohere.com | Trial key for Command models, embeddings and rerank. Never expires, no card needed. | 1,000 calls/month across all endpoints, 20 req/min | Evaluation only; commercial use requires a paid production key. | Jul 2 |
| Hugging Facehuggingface.co | Small monthly inference credit usable across hosted open models. | ~$0.10/month in credits on free accounts | The free credit is symbolic. Real usage needs PRO ($9/mo, $2 in credits) or your own hardware. | Jul 2 |
| OpenAI / Anthropic APIsplatform.openai.com / anthropic.com | OpenAI: a daily free-token allowance if you opt in to sharing your API data (up to ~1M tokens/day on frontier models, ~10M/day on mini). Anthropic: a one-time $5 trial credit. | OpenAI: daily, opt-in only; Anthropic: one-time | OpenAI's free tokens cost you your data. Anthropic's $5 needs SMS verification and is a trial, not a tier. The old OpenAI $5 welcome credit is gone. | Jul 2 |
COD
Coding assistants
5 tracked| Service | What's free | Limits | The catch | Verified |
|---|---|---|---|---|
| GitHub Copilot Freegithub.com/features/copilotno card needed | Completions plus chat and agent mode in VS Code, JetBrains, Neovim and the CLI. | 2,000 completions/mo, 50 chat or agent requests/mo | Auto model selection only. Agent mode eats the 50-request budget very fast. | Jul 2 |
| Antigravity CLIantigravity.google | Terminal coding agent on Gemini models with a personal Google account. Successor to Gemini CLI, which Google shut down for consumer accounts on June 18, 2026. | ~20 agent requests/day, refreshing ~5 h with a weekly ceiling (community-reported) | Google doesn't publish exact numbers and has cut the quota repeatedly (250/day at launch, ~20 now). The CLI shares its credit pool with the Antigravity IDE. | Jul 4 |
| Windsurf Freewindsurf.com | Full editor with unlimited Tab autocomplete. | Daily/weekly quota, ~5 agent (Cascade) actions/day | Pricing was overhauled in March 2026 (credits replaced by daily/weekly quotas); premium models stay gated to paid plans. | Jul 4 |
| Cursor Hobbycursor.com | Limited agent requests and tab completions to try the editor. | Small monthly cap, exact numbers unpublished | Cursor doesn't publish concrete Hobby-plan numbers. It's a demo, not a daily driver; power features need Pro. | Jul 2 |
| Cline / Continue / Aideropen source | The tools themselves, forever. Open source coding agents for VS Code and terminal. | None. Bring your own model | You supply the model. Pair with the free APIs above (Gemini, Groq, Cerebras) for a genuine $0 coding stack. | always |
IMG
Image generation
5 tracked| Service | What's free | Limits | The catch | Verified |
|---|---|---|---|---|
| Bing Image Creatorbing.com/create | GPT-image generation, 4 variations per prompt. Unlimited slow-queue after boosts run out. | 15 fast boosts/week (~60 fast images) | 1024x1024 only, no other sizes. Output licensed personal, non-commercial by default. | Jul 2 |
| Ideogramideogram.ai | Weekly credits, no watermark, best-in-class text rendering inside images. Commercial use explicitly allowed. | 10 slow credits/week, resets Saturday 00:00 UTC | Free generations are public in the community gallery, and the weekly allowance is a fraction of the old 10/day. | Jul 4 |
| Leonardo AIleonardo.ai | Daily token allowance across models, roughly 15 to 35 standard images per day. | 150 tokens/day | Free generations are public by default. | Jul 2 |
| Google Flow (ex ImageFX / Whisk)labs.google | Imagen image generation and limited Veo video with a Google account. | 50 credits/day, no rollover; free credits cover Veo 3.1 Lite/Fast/Quality only | ImageFX and Whisk were shut down in April 2026 and folded into Flow. Output is watermarked (SynthID). | Jul 4 |
| Stable Diffusion / Flux (local)open weights | Unlimited private generation on your own machine. Commercial use OK for most model licenses. | None. Your hardware is the limit | Needs a GPU (8GB+ VRAM for comfort) and some setup patience. Licenses differ per model: FLUX.1 schnell is Apache-2.0, FLUX.1 dev is non-commercial. | always |
AUD
Voice and music
3 tracked| Service | What's free | Limits | The catch | Verified |
|---|---|---|---|---|
| ElevenLabselevenlabs.io | Top-tier text-to-speech, about 10 minutes of audio per month. | 10,000 credits/mo | No commercial use on free; attribution required in anything public. | Jul 2 |
| Sunosuno.com | Full-song music generation, about 10 songs per day on the current model. | 50 credits/day | No commercial use and no downloads: free songs are listen-only inside Suno since the late-2025 Warner deal. Shared (slower) queue. | Jul 4 |
| Whisper (local)open source | Unlimited speech-to-text transcription on your own machine, 90+ languages. | None. Your hardware is the limit | Large models want a decent GPU; small ones run fine on CPU. | always |
VID
Video generation
3 trackedFree video is the stingiest category in AI: daily crumbs, watermarks everywhere, and terms that change fast. It gets re-checked every sweep.
| Service | What's free | Limits | The catch | Verified |
|---|---|---|---|---|
| Klingklingai.com | Daily credit allowance for short clips. | 66 credits/day, ~5 seconds of standard video | Slow queue on free; watermark on output. | Jul 2 |
| Luma Dream Machinelumalabs.ai | Small monthly credit pool in the Dream Machine web and iOS apps. | Monthly credits: web amount unpublished, iOS app lists 250/mo. 720p drafts, no audio | Permanent watermark, non-commercial only, and Luma doesn't publish the web free credit amount. | Jul 4 |
| Runwayrunwayml.com | One-time credit grant to try Gen models. | 125 one-time credits (~25 seconds of video) | Not a recurring free tier: once spent, it's gone. Watermarked. | Jul 2 |
How this ledger works
Free tiers change constantly and silently. Blog posts about them rot in weeks. This page is maintained instead of published: every row carries the date it was last checked by hand against official pricing pages and, where possible, real usage.
What counts as free
- An ongoing allowance you get every day or month, not a one-time trial. One-time credits are always flagged in the catch column.
- "No card needed" tags mean you can sign up without entering payment details.
- The catch column exists because there is always one: training on your data, non-commercial licenses, public galleries, silent limit cuts.
Use this data
The ledger is licensed CC BY 4.0: reuse it freely in articles, apps or research with a link back to getaitools.dev. Machine-readable versions: free-tiers.json (current state), changes.json (every change with before/after values) and llms.txt.
Corrections
Spotted a limit that changed, or a free tier we're missing? Takes 30 seconds, no account needed. Confirmed corrections land in the changelog and you get a permanent spot on the credits page.
Send a correctionQuick answers
Is ChatGPT free?
Yes: as of July 2026, roughly 10 messages every 5 hours on the flagship model, then a smaller model takes over. Opting out of personalized ads can cut the cap further. See the CHT table for details.
Which LLM API has the best free tier?
Google's Gemini API is the most generous ongoing free tier. Groq (1,000 req/day), Cerebras (~1M tokens/day), Mistral (~1B tokens/month) and OpenRouter stack nicely alongside it since every provider's limit is independent.
Does the OpenAI API have a free tier?
Not a traditional one. Opt in to sharing your API data with OpenAI and you get a daily free-token allowance: up to ~1M tokens/day on frontier models, ~10M/day on mini models. The old $5 welcome credit is gone. Anthropic's API offers a one-time $5 trial instead.
Can I build a completely free AI coding setup?
Yes. An open source agent (Cline, Continue, Aider) plus a free API tier (Gemini, Groq, Cerebras) is a real $0 stack. Copilot Free works too if 50 chat requests a month is enough.
Can I use free AI images commercially?
Depends on the tool: Ideogram free allows it, Bing Image Creator and Microsoft Copilot do not. Locally-run Stable Diffusion or Flux is commercial-safe for most model licenses. Always read the row's catch column before shipping anything.
Is there a free AI video generator?
Yes, but it's the stingiest category: Kling gives 66 credits/day (~5 seconds of video), Luma a small monthly credit pool, and Google Flow 50 Veo credits/day. Everything free is watermarked.