The free API models, and what they are actually good for

OpenRouter lists API models at $0 per token. Six of them by name, what each is for, why they are free, and a live run you can do right now with no key and no card.

In this lesson you will

Recognise a free API row and understand why it costs $0
Match six real free models to the jobs they suit
Run a prompt on a free model without an account

The chat tiers in lesson 1 are the free AI everyone knows. The less-known kind is the free API row: a model listed on OpenRouter with an id ending in :free, priced at $0 per million tokens in and out. At the time of writing our catalogue holds 29 of them. They exist because a lab wants usage and feedback, or because a hosting partner is absorbing the cost for a while. That is also why they are rate-limited and can disappear on a Tuesday.

Six free rows from the catalogue, chosen for variety. Context windows are as listed in models.json on 2026-09-12; rate limits are the provider's and change without notice.
Free row (models.json name)	Context window	Good for	Watch out for
Z.ai: GLM 5.2 (free)	256k tokens	General chat, drafting, light coding; the closest thing to a free all-rounder	Rate limits bite during busy hours
OpenAI: gpt-oss-20b (free)	131k tokens	Open-weights OpenAI model; good at instructions and structured output	Smaller than the paid GPT line; not a GPT-5 substitute
Google: Gemma 4 31B (free)	262k tokens	Summaries, rewrites, classification in many languages	No web access, no tools; it is a text model
NVIDIA: Nemotron 3 Super (free)	262k tokens	Reasoning-style tasks and longer answers	Slower; can over-explain when you wanted one line
MiniMax: MiniMax M3 (free)	1M tokens	Very long documents in one go	Free access to a 1M window is rarely fast; expect queues
NVIDIA: Nemotron 3.5 Lightning (free)	1M tokens	Quick answers over big inputs; a good batch classifier	Newer row; check it still exists before building on it

Try one now, with no key

The box below runs on the site's own keyless endpoint, which only ever uses :free rows and the cheapest model in each family, three runs per visitor per day. Edit the text if you like, press Run, and read the cost line at the bottom afterwards.

Try it

Classify three messages on a free model

Three free runs a day on the site key. Add your own key for unlimited runs.

If all four check badges are green, a $0 model just did a job people pay a $20 seat to do by hand. If one is red, read the answer: the usual failure is a polite sentence before the labels, which is a prompt problem (say *nothing else* more firmly), not a model problem.

Knowledge check

Why is a model listed at $0 per token on OpenRouter?

See the live rows

Model catalogue →Filter to $0 rows for today's free list.Playground →The same keyless runs, across every family.

Lesson FAQ

Are free API models worse than ChatGPT?

Usually smaller and slower than the paid flagship, and roughly comparable to a free chat tier on short tasks. GLM 5.2 (free) and gpt-oss-20b (free) handle everyday drafting and classification well; none of them replaces a frontier model for hard reasoning.

Do I need an OpenRouter account to use free models?

For the box in this lesson, no; it runs on the site's keyless endpoint, three runs a day. For your own scripts, yes, you need a free OpenRouter account and key; the free rows still cost $0 through it.

Can I build a product on a free row?

No. They throttle, they may log prompts, and they get withdrawn without notice. Prototype on them, then move to a cheap paid row when it matters; the cheapest paid models cost cents per thousand requests.

Why do the free rows in the catalogue keep changing?

Because the promotions behind them end. The catalogue syncs hourly from OpenRouter, so a row that vanished upstream vanishes here within the hour.

Try one now, with no key

Lesson FAQ

Stop overpaying for AI tools