Free GLM API access: what is actually free.

Quick answer: three real routes, three different catches.

The decision usually reduces to which model you need. If an older Flash-class model is enough, Z.ai's free tier is genuinely free and the right answer. If you need current GLM-5.2 quality, nobody serves that without metering it somewhere, and the honest comparison is between rate-limited free tiers and low per-token prices with a free on-ramp.

• Z.ai's official free tier: the free Flash models (GLM-4.7-Flash and GLM-4.5-Flash at the time of writing) carry no token charges for registered accounts. The catch: they are older, smaller models than the GLM-5 family, and rate limits size them for prototyping.
• Self-hosting: GLM-5.2's weights are MIT-licensed and public. The catch: it is a roughly 750B-parameter Mixture-of-Experts, which puts self-hosting far outside consumer hardware.
• APIsRouter free starting credit: a starting credit with no card required, spendable on the actual GLM-5 family (glm-5.2, glm-5.1, glm-5) plus DeepSeek, Claude, and GPT ids. The catch: it is a starting credit, not an unlimited tier; after it, access is pay-as-you-go.

What Z.ai offers for free, officially.

Z.ai's own pricing docs list two text models as free, GLM-4.7-Flash and GLM-4.5-Flash, with no token charges for input, cached input, or output. Access requires a registered Z.ai account, and third-party guides report request-per-day caps on the free models; Z.ai has revised free-tier details more than once, so verify the current limits in their documentation before building on them. Two boundaries matter. First, the free models are the 4.x Flash line, not the GLM-5 family: GLM-5, GLM-5.1, and GLM-5.2 are paid per token on Z.ai at every tier. Second, free tiers exist to convert prototypes into paid usage, which means the caps are sized to end exactly when a project starts working. That is not a criticism, it is the mechanism, and it is why the practical question is what you migrate to when the cap lands. If your workload genuinely fits a Flash-class model under a daily cap, use the official free tier; it costs nothing and involves no middleman.

What APIsRouter offers: a free start on the real GLM-5 family.

APIsRouter's free route is deliberately simple: start with a free credit, no card required, and spend it on any catalog id, including glm-5.2, glm-5.1, and glm-5. There is no separate free-tier model list and no degraded free endpoint; the credit meters the same models at the same catalog rates as paid traffic. When the credit runs out, access continues as pay-as-you-go top-ups with no subscription, and an idle month costs nothing. To be precise about what this is not: it is not unlimited free GLM. A no-limit free API for a current frontier-family model does not exist anywhere, because someone pays for the GPUs. What a starting credit buys is a real evaluation, current models, real rate behavior, exact per-request token counts in the usage log, without handing over a card or committing to a plan first. The practical pattern: prototype against the credit, watch what your prompts actually cost in the console, and let those numbers decide whether GLM-5 volume pricing or GLM-5.2 quality is the right steady state.

curl https://api.apisrouter.com/v1/chat/completions \
  -H "Authorization: Bearer $APISROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.2",
    "messages": [{"role": "user", "content": "Three test cases for an email validator, as JSON."}]
  }'

The self-hosting route: MIT weights, serious hardware.

The GLM-5 family ships open weights under an MIT license; GLM-5.2's public release landed on June 16, 2026, and the checkpoint on Hugging Face weighs in around 1.5TB. Legally and technically, this is the most unrestricted free GLM there is: no rate limits, no terms beyond MIT, full control of your data. The economics are the catch. A ~750B-parameter MoE with roughly 40B active parameters per token needs multi-GPU server hardware to serve at usable speed, which puts it in the territory of research clusters and infra teams, not laptops. If that is you, the weights are the answer and nothing on this page competes with them. For everyone else, self-hosting a model this size costs more in hardware and operations than years of metered API usage. The middle path some teams take: self-host a small open model for private or bulk workloads, and call the hosted GLM-5 family through an API for the requests that need current quality.

Pay-as-you-go · transparent per-model pricing

Selected models are priced below official list prices. Exact input, output, cache, and per-request prices are shown for each model.

When free stops being the right answer.

Free tiers fail at predictable moments: a daily cap that lands mid-session, a free model that is one generation behind the quality your output needs, a prototype that suddenly has users. The signal to move is never ideological, it is the first time a limit costs you more time than the metered price would have cost money. GLM pricing makes that a low bar. At catalog rates, the GLM-5 family's cost per typical chat turn is a small fraction of a cent, so the workloads that outgrow a free tier usually land at single-digit dollars per month, not subscription-tier bills. The pricing table above shows the exact per-million rates; the token-cost page linked below works the math for real request shapes. Because the starting credit and paid traffic share one endpoint and one key, outgrowing free involves no migration at all: the same base URL, the same model strings, and a balance that tops up instead of a cap that resets.

Estimating what your project needs before you spend anything.

The credit phase is the right time to gather three numbers, because they turn every later decision into arithmetic. Tokens per request: the usage block on each response reports prompt_tokens and completion_tokens exactly, and a dozen representative requests give a trustworthy average. Requests per day: your own product metric. Model mix: what share of traffic genuinely needs glm-5.2 against what passes on glm-5, which one afternoon of side-by-side prompts usually settles. Multiply the three against the catalog rates in the table above and you have a monthly figure before committing anything. For most hobby and early-product workloads that figure lands in single-digit dollars, which reframes the free-API search: the goal was never zero forever, it was zero risk while finding out what non-zero costs. A starting credit answers exactly that.

Which free route fits which situation.

• Learning and small experiments with no quality bar: Z.ai's free Flash models. Genuinely free, official, rate-limited.
• Evaluating the current GLM-5 family before committing: the APIsRouter starting credit, because it runs on the exact ids you would ship with.
• Infra teams with multi-GPU hardware and data-locality requirements: the MIT weights.
• Production workloads of any size: pay-as-you-go metered access; no free mechanism anywhere is designed to carry production traffic.
• Roleplay and chat frontends that need a custom OpenAI-compatible endpoint: the starting credit works there too, since it is the same /v1 surface SillyTavern-style apps expect.

FAQ