“we are about to create a genie that can grant any wish, I think it’s important that the first wishes we in the world as the genie to do benefits the world”
~ sam altman said this in a recent interview he did,
and it’s becoming more clearer that the genie he was talk about was Astra.
OpenAI’s doomsday model lol.
10 PLATFORMS GIVING AWAY FREE AI API CREDITS RIGHT NOW.
no subscription, no invite list. two of these hand you $300 on signup.
between them you'd get gemini, GPT and claude models without paying for a single token.
bookmark this before your next project.
1️⃣ Google Cloud ($300, 90 days): https://t.co/kuxYE9zi00
The biggest single grant on this list. "$300 in free Welcome credit, valid for 90 days," spendable across the whole platform including Gemini models on the Agent Platform.
read this part carefully: the $300 cannot pay for Gemini API in AI Studio, cannot buy third-party models sold as managed APIs, and you cannot attach GPUs while on trial billing.
2️⃣ Oracle Cloud ($300, 30 days + always free): https://t.co/nija6NBZO3
same headline number, tighter window. the real prize is the Always Free tier that keeps running after day 30 with no card charge.
best free infra on the list if you want to self-host models rather than rent an API.
3️⃣ Microsoft Azure ($200, 30 days): https://t.co/OWvtxhulDn
"$200 credit to use on Azure services within 30 days."
this is your GPT route, it works on Azure OpenAI. plus 20+ services free for 12 months and 65+ always free.
the 30 day clock is brutal. unused credit is gone, no extension, no reissue.
4️⃣ AWS (up to $200, 6 months): https://t.co/P2LXJRPx8b
the longest runway here. $100 lands immediately, and you earn up to $100 more by exploring services.
six months, not thirty days. this is your claude route via Bedrock.
5️⃣ Google AI Studio (free forever): https://t.co/JivdQLnhOb
a genuinely free tier, completely separate from the $300 above. free input and output tokens on Gemini 3.6 Flash, 3.5 Flash, Flash-Lite and the embedding models.
5,000 free search-grounded requests a month.
tradeoff: on the free tier your content is used to improve their products.
6️⃣ Cloudflare Workers AI (10,000 neurons/day): https://t.co/UUKkQhF4wO
"a total of 10,000 Neurons per day at no charge," resetting daily at 00:00 UTC. no expiry, no trial clock. ~80 models. the most honest permanent free tier of the bunch.
note: kimi k2.6, k2.7-code and GLM-5.2 need a paid billing method.
7️⃣ Groq (free tier, fastest inference): https://t.co/0Ko92ewOQf
30 requests/min and 14,400 requests/day on Llama 3.1 8B. no card.
the catch nobody mentions: token throughput is only 1.2K-15K TPM, and Llama 3.3 70B drops to 1,000 requests/day. fine for agents, painful for long context.
8️⃣ OpenRouter (one key, every model)
https://t.co/TQ2vKFCkBV
widest model spread for comparison testing, single API key.
the real limit: 50 requests/day on free models. buy $10 in credits once, ever, and it permanently jumps to 1,000/day. that one-time $10 is the highest-leverage spend on this whole list.
9️⃣ Cerebras ($5 credit, 20x faster)
https://t.co/hLLDa0tkR0
"$5 in free credits after making an account," access to all their models. small, but it buys a lot at their speed.
heads up: Cerebras Code Pro and Max are both currently SOLD OUT.
🔟 Mistral (free plan)
https://t.co/R5ups6d9JX
free access to their latest models through chat, search, and agentic coding in your terminal. limited messages and coding sessions, but zero dollars and no card.
add it up. $1,000+ in trial credits across four clouds, plus four tiers that stay free permanently after the trials burn out.
start the always-free ones today. only trigger the cloud trials when you have a real project ready, because every clock starts the second you sign up.
save this before you pay for another API key.
Follow me and turn on 🔔 post notifications so you never miss the next one.
10 GITHUB REPOS THAT WILL SAVE YOU THOUSANDS IN SAAS SUBSCRIPTIONS.
all completely free and open-source. every one of them replaces a tool people are paying monthly for right now.
bookmark this for later.
1️⃣ palmier pro (palmier-io/palmier-pro) ⭐ 13k
swift-native mac video editor built around AI. generate footage with seedance, kling, or nano banana pro directly inside the timeline, and connect claude code / codex / cursor over MCP so your agent edits alongside you.
replaces: premiere pro ($22.99/mo), descript ($24-50/mo), capcut pro.
2️⃣ strix (usestrix/strix) ⭐ 48k
autonomous AI hackers that actually run your app, find the vulnerabilities, and hand you a validated fix. plugs into github actions so every PR gets scanned before merge.
replaces: burp suite pro (~$475/yr) and pentest-as-a-service retainers that start in the five figures.
3️⃣ PaddleOCR (PaddlePaddle/PaddleOCR) ⭐ 87k
turns any PDF or image into clean markdown/JSON. PaddleOCR-VL is a 0.9B model hitting 96.3% on OmniDocBench, handles tables, formulas, seals, handwriting, 100+ languages.
replaces: mathpix, ABBYY, adobe acrobat pro OCR, AWS textract and azure document intelligence per-page billing. unlimited pages, runs on your own CPU.
4️⃣ Scrapegraph-ai (ScrapeGraphAI/Scrapegraph-ai) ⭐ 29k
you describe the data you want in plain english, it builds the scraping pipeline with LLM + graph logic. works on live sites and local HTML/XML/JSON/markdown.
replaces: apify, octoparse, https://t.co/RWgVsVJNdy, firecrawl credits. no per-request pricing.
5️⃣ OmniVoice Studio (debpalash/OmniVoice-Studio) ⭐ 9.5k
zero-shot voice cloning, real-time dictation, full video dubbing, audiobook generation. 646 languages, no accounts, no API keys, everything on your machine.
replaces: elevenlabs ($99-330/mo), descript overdub, rask. this is the biggest single line item most creators are paying.
6️⃣ LTX-2.3 (Lightricks/LTX-2) ⭐ 8.5k
first DiT audio-video foundation model, generates synchronized sound and video in one pass. open weights plus a LoRA trainer, so you can fine-tune it on your own footage.
replaces: sora, runway ($95/mo unlimited), kling credits. generate on your own GPU instead of buying render credits.
7️⃣ LongCat-Video (meituan-longcat/LongCat-Video) ⭐ 6.6k
13.6B model from meituan. text-to-video, image-to-video, and video-continuation in one architecture, so it produces minutes-long clips without color drift. the avatar 1.5 release does audio-driven talking heads with whisper-large lipsync.
replaces: heygen and synthesia ($29-89/mo per seat) for avatar video, plus long-form gen credits.
8️⃣ opencode (anomalyco/opencode) ⭐ 193k
the open source coding agent. terminal-native, works with claude, codex, gemini, GLM, kimi, or any local model you point it at.
replaces: cursor ($20/mo), github copilot ($10-39/mo), devin. bring your own key and the tool itself costs nothing.
9️⃣ OpenPencil (open-pencil/open-pencil) ⭐ 7.5k
AI-native design editor that reads and writes native .fig files, so you can leave figma without leaving your work behind. 90+ AI tools for building nodes, plus a headless CLI and MCP server for agents.
replaces: figma ($16-55 per editor per month). that is per seat, every month, forever.
🔟 OpenReply (diwenne/openreply) ⭐ 890
someone comments a keyword on your reel, they get a DM with your link a second later. tracked links, click stats, follow gating, all through meta's official private replies API.
replaces: manychat. same core feature, self-hosted, no seat limits and no plan caps.
add it up. that is one coding agent, one pentester, one OCR engine, one scraper, one voice studio, two video models, one editor, one design tool and one DM automation, for $0 in subscriptions.
save this list before you renew anything.
Follow me and turn on 🔔 post notifications so you never miss the next one.
🚨 Opus 5 vs GPT-5.6 (sol)
I tried the exact same prompt with GPT-5.6 (sol), and its output was even better.
GPT-5.6 took less time and cost significantly fewer tokens, although opus 5 still had better audio quality.
i’ll share the prompt i used in the comments.
$500 challenge to whoever who can perform best with highest RR.
ends this sunday
Dm https://t.co/wihs4i7JaG to join our discord today for free
limited spots - Anyone can join and win :)