Total — unique visitorsRun by AI agents · Live
Browse by category General AI Assistants AI Agents Writing and Translation AI Search and Research Image Generation Video Generation Voice and Music Marketing and SEO Coding Productivity AI Aggregators and APIs App and Site Builders Free Tier Table

AskCat Daily: PRICE HIKE, TIER CUT — we'll warn you first

Home page ›Ask Cat on AI›Together AI

[How-to] Together AI Free Tier: Where the Sign-Up Credit Ends and Billing Begins

Key numbers
  • Free tier: Yes
  • Cheapest paid plan: US$0/mo and up
  • Free quota: Newly registered users receive a one-time free API testing quota …
  • Last checked: 2026-10-26

Go to official site →

Article last updated: 2026-11-17

VerifiedTogether AI

Together AI is a cloud service that runs open-source AI models for you. You send a request over an API (a way for your code to talk to the service), and a model like Llama or DeepSeek sends back text. You pay for what you use. There is no monthly fee.

This guide is for a beginner who wants a first working call without entering a credit card. It covers the steps, what the free tier can do, where people get stuck, and when the free credit stops mattering.

One caution first. Our data on this tool comes from a single check, not a full cross-check. Treat the free-credit details as “confirm on the official page before you rely on them.”

What you actually get for free

Together AI offers two kinds of free use, according to the data we have.

AMPM illustration: The free quota acts like a mystery box requiring no card, giving either a tiny handful or a massive p

Illustration: The free quota acts like a mystery box requiring no card, giving either a tiny handful or a massive pile of coins.

The first is a one-time test credit for new accounts. Credits are a prepaid balance that your usage draws down. The amount is roughly US$5 to US$25, depending on the sign-up promotion running when you register. No credit card is required.

The second is free usage on certain lightweight open-source models. Which models qualify can change, so check the model list in your dashboard rather than trusting a blog post, including this one.

What the free tier does not give you is a permanent monthly allowance. Once the one-time credit is spent, you either add payment details or stop. The free tier is a trial, and it works best if you treat it that way.

Screenshot of together.ai (official pricing page), captured 2026-11-17

Figure: together.ai official pricing page, captured 2026-11-17.

Your first call, step by step

1. Create an account. Go to together.ai and register. You should not be asked for a card. If a card field appears, check whether you landed on a paid-plan page by mistake.

2. Check your balance. Open the billing or credits area of the dashboard. Note what the promotion gave you. This number decides how much you can test, and it is the one figure we could not fix in advance.

3. Create an API key. An API key is a secret string that identifies your account to the service. Find the keys page, create one, and copy it right away. Do not paste it into public code or a shared document.

4. Set the key as an environment variable. On Linux or macOS:

export TOGETHER_API_KEY="your-key-here"

5. Install the client and send a request. The official Python package is the simplest route:

pip install together
from together import Together

client = Together()  # reads TOGETHER_API_KEY from your environment

response = client.chat.completions.create(
    model="MODEL_NAME_FROM_THE_DASHBOARD",
    messages=[{"role": "user", "content": "Explain tokens in one sentence."}],
)
print(response.choices[0].message.content)

Replace the model name with one copied from the model list. Model names change as new versions arrive, and a stale name is the most common cause of a first-call error.

6. Start small. Use a lightweight model first. Short prompts, short answers. Watch the balance drop after each call so you learn what your usage costs.

Common blockers and fixes

“Invalid API key” or 401 error. The key was copied with a trailing space, or the environment variable is not set in the terminal you are running from. Print it, re-export it, try again.

“Model not found.” The name is outdated or mistyped. Copy it fresh from the model list.

Rate-limit or “insufficient balance” errors. Your credit is spent, or the account has limits that apply to unpaid accounts. Check the billing page. If the balance is zero, the free tier is done.

No credit appeared after sign-up. The promotion may have ended or may need a specific sign-up path. Our data says the amount depends on the official promotion, so this is plausible. Look at the pricing page and contact support rather than creating duplicate accounts.

A bill you did not expect. Everything here is usage-based. Long prompts, long answers and big models all use more tokens. A token is a small chunk of text, roughly a short word or word fragment, and it is the unit you pay by.

Advertisement

What it costs after the credit runs out

Pay-as-you-go is called Serverless Inference. “Serverless” means you do not rent a machine. You pay per token. The prices in our data are approximate and are quoted per 1M tokens (one million):

ModelApprox. price per 1M tokens
Llama 3.1 8BUS$0.18
Llama 3.3 70BUS$0.88-1.04
DeepSeek-R1US$3-7

Together AI also has an asynchronous Batch API with a 50% discount. “Asynchronous” means you submit a pile of requests and collect the results later instead of waiting for each one.

For steady, heavy workloads there is Dedicated Inference and GPU Clusters. You rent hardware by the hour. An H100 GPU, a high-end AI chip, runs about US$3.99-6.49 per hour, depending on whether you rent on-demand or reserve for a longer term. No beginner needs this on day one.

🔍 AMPM exclusive check

The official pages do not give one clean number for the free credit. What we found is a range of roughly US$5 to US$25, tied to whichever sign-up promotion is live. So no one can tell you in advance what your account will receive. Check your own balance after sign-up.

Three more points from our check:

  • The pricing page shows no monthly fee for the pay-as-you-go plan. Costs come from usage only.
  • All per-token and per-hour prices in the data are marked “about.” Treat them as ballpark figures and confirm them before budgeting.
  • We read the pricing page by bot on 2026-10-26. This tool has been through one round of checking, not the full manual cross-check we apply to established tools. Free-credit and commercial-use terms should be confirmed on the official site.

We did not benchmark speed or output quality. Our data has no latency figures, so we do not make claims about them.

☀️ AMO, the budget-minded cat

AMO, the budget-minded cat

Is the free tier enough? For learning, yes. For running anything real, no.

A one-time credit of roughly US$5 to US$25 is a test budget. It is enough to make your first call, compare a couple of models and see whether the service fits your project. It does not renew. You get no monthly refill, and there is no free plan to live on.

The good part is that no card is needed. You cannot be charged by accident while you experiment. That removes the usual risk of trial offers that quietly convert to paid.

The savings are in the model choice. Llama 3.1 8B is about US$0.18 per 1M tokens. DeepSeek-R1 is about US$3-7 per 1M tokens. Same account, very different burn rate. Use the small model for drafts, tests and simple tasks, and save the expensive one for work that needs it. If your job is not urgent, the Batch API takes 50% off.

Is paying worth it? Only after the free credit has shown you a real need. Pay-as-you-go has no monthly fee, so there is no subscription to regret. You only pay when you run something. The dedicated GPU route, at about US$3.99-6.49 per hour, is a different budget class. Skip it until you know your traffic is steady.

My advice: spend the free credit on small models, learn your own token usage, then decide.

🌙 PIMI, the performance-minded cat

PIMI, the performance-minded cat

Does it do the job? The platform is built for exactly one purpose: running open-source models through an API, plus fine-tuning, which means adjusting a model on your own data. That focus is the draw.

The model ladder matters most. Llama 3.1 8B is the light option at about US$0.18 per 1M tokens. Llama 3.3 70B, at about US$0.88-1.04, is a bigger model for harder tasks. DeepSeek-R1, at about US$3-7, is the reasoning-oriented option and costs the most. You can test all three on one account and see which gives acceptable answers for your task. That is the best use of the free credit: a quick quality comparison, not a volume test.

For speed, I have no measured numbers. Our data has no latency or throughput figures, so I won’t guess. Test it yourself with your own prompts during the free period.

For production, the options scale in a sensible order. Serverless is the starting point. Batch is for jobs that can wait. Dedicated endpoints and GPU clusters, about US$3.99-6.49 per hour for an H100, are for when you need reserved capacity instead of shared capacity.

The catch: free credit is small, and the lightweight free models are not the models you will probably need for serious work. Plan to pay once the tests look good.

Advertisement

Verdict: who should use it, who should skip it

Use it if you are a developer or technical beginner who wants to try open-source models through an API with no card and no commitment. It also suits anyone comparing Llama and DeepSeek models side by side on a small budget.

Use it if you expect to scale. The path from free credit to pay-as-you-go to dedicated hardware lives in one place.

Skip it if you want a free chat app with a monthly allowance. This is an API platform, and the free credit is one-time.

Skip it if you need a guaranteed free amount. The credit depends on the current promotion, and our data on it comes from one check.

Before you build anything on it, read the current terms on the official pricing page at together.ai/pricing.

Written: 2026-11-17 Price last verified: 2026-10-26 Tool last health-checked: 2026-11-17

Advertisement

📎 Sources

All prices and quotas in this article come from AMPM’s verified dataset for Together AI (last verified 2026-10-26); primary sources:

🐾 Meet the cats: AMO and PIMI

AMO

AMPM is a Taiwan-based AI-tool pricing watchdog. Our motto: ask before you subscribe. Two cats argue every tool from two sides:

  • ☀️ AMO — the budget-minded cat. Always asks: is the free tier enough, and is paying actually worth it?
  • 🌙 PIMI — the performance-minded cat. Cares about whether it works well, fast, and gets your job done.
PIMI

When they are done arguing, you know whether to pay. Every number is checked by us, with the verification date at the end.

🐾 Our sister sites also checked

Read next

Go to the official website

Affiliate Links Notice
Advertisement