A recruiter rewrites the same outreach email four times a week. A hiring manager builds interview questions from scratch for every role. An engineer is asked whether any of this can be automated, and the budget is zero. Free LLM access makes that first experiment cheap.
Cheap is not the same as safe or permanent. This guide covers which small hiring-adjacent helpers are worth prototyping, what to check before you rely on a free tier, and how to keep your code simple enough to move later.
Three small prototypes worth building first
Start where mistakes are cheap and a person reads every output.
- Outreach drafting: feed in a role brief and a few notes about a candidate’s public work, and get a first draft a recruiter edits before sending.
- Interview question sets: feed in a job description and the competencies that matter, and get structured questions with a short note on what a strong answer covers.
- Résumé summaries: condense a résumé into a neutral paragraph a reviewer can scan. Summarize, do not rank.
Each task has a clear input, a readable output and an obvious test of quality: would a colleague send this? That makes them good material for comparing models before you commit to one.
Keep a person in the decision
Anything that touches who gets an interview deserves extra care. Rules on automated hiring tools differ by country and region, and they change, so treat the points below as questions for your HR and legal colleagues, not as answers.
- Does a person make every advance-or-reject decision?
- What candidate data leaves our systems, and are we comfortable with the provider’s terms for it?
- Who checks outputs for uneven treatment of similar candidates?
- Do candidates need to be told that software assists the process?
Prototype with sample or anonymized data until those questions have answers. A free tier is a fine place to learn. It is a poor place to process real candidate records.
What to check before you trust a free tier
Free offers expire quietly. Before building on one, check four things on the provider’s own pages.
- Status: does the model page say it is active? Roundups age fast. When one published list of free models was compared with the live pages in November 2026, two entries were marked as sunset and a third page no longer existed.
- Limits: free models carry a rate limit. Read the current label on the page and add your own request caps instead of trusting a number remembered from an article.
- Terms: is the free tier meant for testing, or may you run a product on it?
- Path to paid: which plan do you move to, and how does it bill?
If you are still building a shortlist, ZenMux’s roundup of free LLM API options developers can actually use is one starting list. As with any roundup, confirm each model’s current status on its own page using the checks above.
Two free models listed today, side by side
In November 2026, ZenMux’s model pages list two free models marked Active: GLM 4.6V Flash (Free) and GLM 4.7 Flash (Free), both from Z.AI. The table summarizes how each page describes them and where each might fit in a hiring-tool prototype.
| Model | Listed status | Context shown | Described strengths | Prototype idea |
|---|---|---|---|---|
| GLM 4.6V Flash (Free) | Active, Free, Rate Limit label; figure: Not listed | 200K | Multimodal understanding with a reasoning toggle; native function calling | Pulling fields from image-based documents, using sample data only |
| GLM 4.7 Flash (Free) | Active, Free, Rate Limit label; figure: Not listed | 200K | 30B-class model tuned for agentic coding, long-horizon planning and tool collaboration | Drafting outreach and generating interview question sets |
Prices and specs as listed on official pages in November 2026; check the live pages before deciding.
A model page describes a model. It does not promise quality on your task. Run the same sample tasks through each one and compare the results, as the next section suggests.
Keep the plumbing swappable
A prototype should be easy to throw away. Many gateways, ZenMux among them, expose a unified, OpenAI-compatible endpoint, so moving a prototype often means changing a base URL and an API key rather than rewriting logic. ZenMux model names follow a provider/model-name pattern, which makes the model one more configuration value.
Three habits keep you flexible, with no lock-in to any single provider:
- Keep the base URL, key and model name in configuration, never in code.
- Wrap model calls in one small function so the rest of the app never touches a vendor-specific SDK.
- Keep 15 to 20 anonymized sample tasks and run every candidate model through them. The comparison takes minutes and replaces opinion with evidence.
One caveat: ZenMux is a hosted service, so teams that need a self-hosted or on-premises gateway should look at other options.
Make usage observable from day one
Free quotas disappear quickly when three prototypes share one key. Create a separate API key for each prototype and review the request logs. ZenMux’s documentation describes logs you can filter by API key, model, provider, request ID and finish reason, with tokens, cost, latency and throughput recorded for every request.
Finish reasons are worth reading. A value such as max_tokens or length suggests an answer was cut off, a likely reason for half-finished interview questions. Whatever gateway you choose, look for the same basics: tokens, latency and finish reason per request.
Moving from prototype to production
The day a helper becomes part of a team’s weekly routine, it stops being a prototype. Read the plan terms again. ZenMux’s quickstart describes Pay As You Go as prepaid, billed per token and meant for production and commercial products. Its subscription plan is for personal development and learning, and production use is not allowed on it.
Also confirm which account type can call a given free model through the API. Do not assume it from the model page. Then re-run your sample tasks on the model you plan to pay for, and ask your HR and legal colleagues to revisit the questions above before real candidate data enters the system.
Frequently Asked Questions
Can I use AI API for free?
Yes, within limits. Some providers and gateways list models at no cost, usually with a rate limit and a status that can change. In November 2026, ZenMux’s model pages list GLM 4.6V Flash (Free) and GLM 4.7 Flash (Free) as active. Treat free access as a way to test, and read the model page for current limits before building on it.
How do you use AI for recruiting?
Start with low-risk writing and summarizing tasks: outreach drafts, interview question sets and neutral résumé summaries. A person should read every output and make every hiring decision. Use sample or anonymized data while you prototype, and ask your HR and legal colleagues which rules and candidate notices apply where you hire.
Which LLM API is totally free?
None can be assumed free forever. Free offers are usually limited, and models get retired, so a list that was right last quarter may be wrong today. Check each model’s status and rate-limit label on its own page, note the terms for production use, and keep your code provider-agnostic so a change is a configuration edit.
What are AI tools for recruitment?
They range from ready-made products for sourcing, screening and scheduling to small in-house helpers built on an LLM API. Small teams often prototype the in-house kind first: an outreach drafter, a question generator or a résumé summarizer. Whichever route you take, ask where candidate data goes and who reviews the output.
Conclusion
Free LLM tiers are a good way to learn what AI can do for a hiring workflow. Keep the first prototypes small, keep a person in every decision, check status and limits on the live page, and keep your code ready to move. Do that and the step from experiment to a paid, production setup stays simple, reliable and transparent.