GitHub Trending on August 18: MoneyPrinterTurbo leads, plus AWS credits and free AI access (2026)

The fastest mover on today's GitHub board is MoneyPrinterTurbo, up 1,189 stars in the daily view. Cordis is close behind at 957, while Strix adds 598. The useful question is what each project lets you test before its star count tempts you into a bigger commitment. 1

The free section follows the same rule. AWS is offering new accounts $100 in credits, OpenRouter's live catalog includes a zero-cost 1-million-token model, and OpenAI's academic program is taking waitlisted applications. Each has a boundary that matters before you claim it.

The board, sorted by stars gained

Rank	Repository	Stars gained	Total stars	Language	What it is and the first useful move
1	harry0703/MoneyPrinterTurbo	+1,189 1	106,522 1	Python	An automated workflow that generates HD short videos from a topic or keyword. Try one controlled input and inspect the output before treating it as a production pipeline. 2
2	cordiverse/cordis	+957 1	5,743 1	TypeScript	A meta-framework for spatiotemporal composability. Its README says the API is under active development and may change without notice, so keep the first prototype isolated. 3
3	usestrix/strix	+598 1	54,388 1	Python	An open-source AI penetration-testing tool for finding and fixing application vulnerabilities. Run it only against systems you own or are authorized to test. 4
4	agalwood/Motrix	+344 1	53,195 1	TypeScript	A full-featured download manager. The sensible first check is whether its desktop workflow fits your operating system and download habits. 5
5	santifer/career-ops	+218 1	64,875 1	JavaScript	A local AI job-search tool that scans listings, scores them with an A-F rubric, tailors a CV, and tracks applications inside an AI coding CLI. Test it on a small set of listings before trusting its ranking. 6
6	akitaonrails/ai-memory	+207 1	2,233 1	Rust	A long-term memory layer for agent coding CLIs and handoffs between different agent vendors. Start with one handoff scenario and check what context is actually retained. 7
7	mukul975/Anthropic-Cybersecurity-Skills	+198 1	28,552 1	Python	A library of 817 structured cybersecurity skills mapped to six frameworks and 29 security domains, with Apache 2.0 licensing. Check its compatible-platform list before adding skills to an agent workflow. 8
8	AlexsJones/llmfit	+198 1	32,402 1	Rust	A command-line tool for finding which of hundreds of models and providers can run on your hardware. Use it before downloading a model, when memory or compute is the real bottleneck. 9

The daily page also shows Cordis as the biggest follow-up story after MoneyPrinterTurbo. Its current +957-star jump is worth reading as a change in momentum, while its own warning about API stability is still the reason to keep it out of a core dependency. 1 3

What the numbers do not tell you

MoneyPrinterTurbo is the attention grabber. Its description points to automated short-video generation, so the quickest useful test is a single prompt-to-video run. The star count says nothing about output consistency, model cost, or how much manual cleanup the result needs. 2

Strix has a clear permission boundary. Security tooling is useful only when the target and the test are authorized. Keep a scan inside a disposable lab or an application you control, and treat findings as leads for review rather than automatic proof of a vulnerability. 4

llmfit answers a hardware question, not a quality question. It can help narrow the model list to what a machine can run. You still need a small task set to compare latency, tool use, and output quality on your own workload. 9

Free developer paths to claim or try

1. AWS Free account plan: $100 now, up to $100 more

AWS says new customers receive $100 in credits after creating an account, regardless of whether they choose the Free account plan or the Paid account plan. Customers can earn up to another $100 by completing listed activities, and the service also includes more than 30 always-free services with monthly usage allowances. 10

Who it fits: a developer building a proof of concept who has never opened an AWS account. The Free account plan is designed for experimentation and can avoid charges for up to six months while the plan is active. 10

How to claim: open AWS Free Tier, choose Create free account, and select the Free account plan during signup. 10

The catch: the Free account plan ends after six months or when the credits are fully used, whichever comes first. It also excludes some services that could consume the allowance quickly and does not cover hardware purchases. A production workload belongs on a paid plan with its normal pay-as-you-go terms. 10

2. OpenRouter: a free 1-million-token model to test

OpenRouter's current model catalog lists nvidia/nemotron-3.5-lightning:free at $0 for input and output tokens, with a 1,000,000-token context window and a 65,536-token maximum completion at the top provider. The catalog describes it as a text-in, text-out NVIDIA Nemotron mixture-of-experts model with 3B active parameters out of 30B total. 11

OpenRouter's official :free documentation says you can append :free to a model ID to access its free variant. Free variants can have different rate limits and availability from paid versions. 12

Who it fits: a builder who needs a no-token-cost endpoint for a coding experiment, agent prototype, or long-context test. The model's large context window makes it a sensible candidate for testing repository-sized prompts, while the result still needs evaluation on your own tasks. 11

How to claim: create an OpenRouter API key, use the exact model ID nvidia/nemotron-3.5-lightning:free, and start with a small test request. The current OpenRouter guide says a free account gets 50 free-model requests per day and 20 per minute; adding at least $10 in credits raises the daily ceiling to 1,000 while the minute limit remains 20. 13

The catch: a zero token price is still a bounded service. Rate limits, availability, provider changes, and failed requests that consume quota can matter more than the headline context window. Keep a fallback and do not make a production SLA depend on a free endpoint. 12 13

3. ChatGPT for Academic Researchers: join the waitlist

OpenAI's program is aimed at researchers at selected academic institutions. It plans to provide 100,000 researchers with free access through 2027, beginning with a 10,000-researcher cohort. Participants receive access across ChatGPT, ChatGPT Work, and Codex, with expanded deep research, higher usage limits, and larger context windows. 14

Who it fits: scientists, mathematicians, and engineers whose work is affiliated with a qualifying degree-granting research institution. The application requires institutional-affiliation verification plus information about active research and intended scientific use. 14

How to claim: join the waitlist and check the eligible-institutions list. OpenAI's August 10 update says more than 13,000 first-wave applications represented up to 65,000 researcher seats, so the initial 10,000 seats will be selected by lottery among eligible applicants. 14

The catch: this is an eligibility-based waitlist, not an instant free account. Approved researchers can invite up to four collaborators from their institution, but each collaborator must verify affiliation and counts toward the program's account total. 14

What to claim and test first

Claim AWS first if you are new to AWS and have a proof of concept that fits the six-month Free account plan. Treat the credits as a test budget, not a permanent cloud subsidy. 10
Try OpenRouter's Nemotron model if you want a free long-context API experiment. Log request counts and failures against the 50-per-day and 20-per-minute limits before you build an agent loop around it. 11 13
Join the academic waitlist if you work at an eligible research institution and can describe an active scientific use. The program's free access includes Codex, which makes it relevant to research software and reproducible analysis as well as writing. 14
Star or clone selectively: MoneyPrinterTurbo for automated video generation, Strix for authorized security testing, Cordis for framework exploration, and llmfit for matching models to local hardware. Use the README boundary before you use the star count as a reason to adopt anything. 2 3 4 9

Today's split is practical: star growth tells you where to look, while the README, quota table, or eligibility page tells you what you can safely do next. The free offers are useful because their limits are visible; those limits are part of the offer, not a footnote to ignore.