free-ai-tools: a curated index of free LLM tiers, with an expiry date on every entry
Curated list of free and low cost AI tools, LLM APIs, IDEs, agents, and infrastructure for building real AI apps
At a glance
- What is it?
- ShaikhWarsi/free-ai-tools is an MIT-licensed Markdown list of free and low cost AI APIs, IDEs, agents and infrastructure, plus a small TypeScript site for browsing it. Its value is the pricing and rate-limit detail; its weakness is that the same detail decays on a schedule the repo cannot control.
- Who is it for?
- Adopt it if you are choosing a free or low cost LLM tier for a prototype and want rate limits, commercial-use notes and price-per-million-token figures in one place, or if you want the RAG, embedding, vector hosting and architecture-pattern sections as a starting checklist. Skip it if you need a package to install, a runtime, or anything with a stability guarantee: this is a Markdown document and a TypeScript browsing site, not infrastructure.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 37 days ago.
- What is it written in?
- Mainly TypeScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
The problem is not finding AI tools, it is finding free tiers that survived the last pricing change
The README opens with a diagnosis of the category it competes in: most AI tool lists are, in its words, outdated, filled with affiliate links, and missing production-critical details such as rate limits, commercial use and architecture patterns. The target reader is named directly: builders who want to ship AI features this week, and the stated goal is to help developers build AI apps without paying $200/month. That framing matters because it sets the inclusion bar. The README says it focuses on tools developers actually use in production, generous free tiers with no five-requests-then-paywall pattern, production-capable models, and real infrastructure such as APIs, hosting and vector databases rather than chatbots. It also states a contribution rule: pull requests should not add the author's own paid or expensive personal projects. The audience is therefore a developer picking a model provider or an IDE for a side project, a prototype, or the early stage of a product, and who cares about the difference between 1,500 requests per day and a trial credit that expires.
The artefact is a README with a comparison spine, not a library
There is no package to install and no API to call. The repository is a curated list, and the primary language flag of TypeScript refers to the companion site at FreeAiToolsList.vercel.app, added in the 2026-04-12 update according to the changelog. The substance lives in the Markdown table of contents, which is organised as a decision sequence rather than an alphabetical dump. It starts with a Quick Comparison of free LLM API providers, splits providers into fully free and trial-credit categories, then separates IDEs and CLI coding tools into those with pro-grade models and those with basic models. From there it moves into API providers for AI coding tools, a paid tiers comparison, local models, and a section called free-coding-models CLI. The second half is infrastructure-oriented: recommended stacks, realtime and streaming APIs, speech, image and video generation, browser automation, cheap vector DB hosting, common AI architecture patterns, model price comparison, best models by use case, rate limit comparison, commercial use summary, RAG stack tools, free embedding APIs, AI hosting and GPU providers, evaluation tools and structured output tools. That ordering is the actual design: pick a provider, then check whether the licence and rate limits allow the thing you want to build.
The maintenance model is dated changelog entries, and the README admits some rows are unverified
A list like this lives or dies on freshness, and the repository handles that explicitly. The changelog is dated and specific. The 2026-06-25 entry describes a model verification and name alignment pass that migrated old placeholders to Claude Fable 5, Claude Opus 4.8 and GPT-5.5 architectures. The 2026-06-16 entry is the most instructive: it added OpenCode, AWS Kiro and a Xiaomi MiMo token plan, and removed items from the free LLM providers section on the grounds that they were no longer free or only partly free, naming Cohere as non-commercial only, GitHub Models as Copilot-required, SambaNova and Hyperbolic as trial-only, HuggingFace at roughly $0.10 per month, Vercel at $5 per month, and Mistral Codestral, Together AI, iFlow and the Perplexity API. That same entry refreshed pricing for Windsurf, Trae, Qoder and GitHub Copilot. The README also carries a warning that early 2026 model tier changes pushed flagship reasoning models to paid tiers, leaving free tiers with lighter versions, and that entries marked with [verify] need confirmation. Read that literally: the maintainer is telling you the list is a map with known blank spots, not a source of truth. The most recent push timestamp is 2026-08-09, so the changelog's latest dated entry is not the latest activity.
What the pricing tables actually contain, and why the numbers rot
The concrete payload is price and quota. The README advertises cheapest AI APIs at $0.08 to $0.50 per 1M tokens, and the changelog records a specific example: Xiaomi MiMo V2.5 Pro cut prices by 99% on May 26, to $0.435 input and $0.87 output with $0.0036 cache pricing. It also records the billing model shifts that break naive cost estimates. Windsurf switched to a quota model on March 18 with Pro at $20, Teams at $40 and a new Max tier at $200. Trae moved to a five-tier token system on February 24 with Lite at $3, Pro at $10, Pro+ at $30 and Ultra at $100. Qoder's 50% launch promotion ended April 30, leaving Pro at $20, Pro+ at $60 and Ultra at $200. GitHub Copilot moved to usage-based billing called GitHub AI Credits on June 1, with a Max tier at $100. The Gemini CLI entry was updated to note that 3.1 Pro is paid-only while 3 Flash is the free tier at 1,500 requests per day. These are the rows worth reading, and they are also the rows most likely to be wrong by the time you read them, because every one of them is a vendor decision the repository can only observe after the fact.
The free-tier trap the README warns about: lighter models behind unchanged names
The warning block is the most useful paragraph in the repository. It states that major providers restricted flagship reasoning and pro models to paid tiers, and that free tiers now receive optimized or lighter versions, listing GPT-5.5 Instant, Claude Sonnet and Haiku, and Gemini Flash as the free substitutes. The practical consequence is a failure mode that has nothing to do with the list being wrong: a developer benchmarks a task on a paid model, then ships it against the free tier of the same family and finds the quality has changed, because the name on the free tier points at a different model than the one tested. The repo's own contribution guidance is a second, softer constraint. The note asking users not to abuse these services, because otherwise we might lose them for everyone, is a recognition that a widely circulated list of free tiers changes the load those free tiers receive. If you are the reason a provider tightens a limit, the list is the mechanism. That is not a reason to avoid it, but it is a reason to treat every free tier as capacity you do not control and to keep a paid fallback path configured rather than assumed.
Alternatives: a general awesome list, a marketing directory, or the provider's own page
The README positions itself against three named categories: awesome-ai as a general list, ai-collection as marketing-focused, and toolify as affiliate-heavy. The real difference is scope discipline. A general awesome list optimises for coverage, so it accumulates entries without rate limits, commercial-use notes or price-per-million-token figures, and you end up opening twenty tabs to answer one question. An affiliate directory optimises for click-through, which is why the README's contribution rule bans adding your own paid projects. The honest alternative for a single decision is the provider's own pricing page, which is authoritative and current but tells you nothing about the other nineteen options. The repository's value is the comparison axis it imposes: free versus trial credit, pro-grade versus basic models, commercial use summarised in one table. If you already know which provider you want, the list adds nothing you cannot get faster from the vendor. If you are still choosing, the side-by-side rate limit and commercial use tables are the reason to open it.
Licence, contributions and what a fork actually costs you
The repository is MIT licensed, which permits reuse and modification with attribution and without warranty, though this is not legal advice and the underlying tool listings carry their own terms that MIT does not touch. The contribution path is documented in CONTRIBUTING.md, and the README frames contributions as welcome, with the specific request that pull requests not add paid or expensive personal projects. That is a low-friction model: no build step to run before a content change, and the site is deployed separately. The maintenance cost is the interesting part. Keeping this accurate means tracking billing changes at every listed provider, which is why the changelog reads as a series of reactive entries rather than a schedule. A single maintainer doing verification passes every few weeks will always lag vendor announcements by days or weeks, and the [verify] markers are the visible residue of that lag. If you fork it for internal use, the cost is not the code, it is assigning someone to re-check the rows you actually depend on, and the repository gives you no tooling for that beyond the changelog format itself.
Editorial conclusion
Adopt it if you are choosing a free or low cost LLM tier for a prototype and want rate limits, commercial-use notes and price-per-million-token figures in one place, or if you want the RAG, embedding, vector hosting and architecture-pattern sections as a starting checklist. Skip it if you need a package to install, a runtime, or anything with a stability guarantee: this is a Markdown document and a TypeScript browsing site, not infrastructure. Before you commit a tier to a production path, open the provider's own pricing page and confirm the model name, the free request allowance and the commercial-use terms, because the README itself marks some entries [verify] and logs provider billing changes by date.
Community notes