Ertas Pro Plan Review ($50/mo): 200 Credits, 100 GB

The Ertas Pro plan is $50 per month. It is built for power users and small teams who are training models continuously rather than occasionally, including consultants and agencies running work for several clients.

This is an honest breakdown of what the plan includes, who it is designed for, how the ROI math works, and when it is not enough.

What's Included in Pro

200 credits per month. Double Builder's 100. Depending on dataset size and model, that is roughly 12 to 25 training runs a month, which covers a monthly retraining cycle across a handful of active models with headroom for the experiments that do not work out.

100 GB of model artifact storage. Every completed run leaves a LoRA adapter and, if you enabled it, a GGUF export in the Hub. A 7B Q4 GGUF is about 4 to 5 GB and a 14B Q4 GGUF is about 8 to 9 GB, so 100 GB holds somewhere between 12 and 20 exported models alongside their adapters.

5 GB of dataset storage. Separate from the model quota and account-wide, covering uploaded JSONL, CSV, parquet and text files plus anything mirrored in from Hugging Face.

Priority GPU queue. Your runs are scheduled ahead of Free, Lite and Builder runs when the queue is busy, which is the difference between starting a retrain now and starting it after lunch.

Preview access to new features. Pro accounts get new Studio capabilities before they reach the lower tiers.

Higher-tier GPU access. Better hardware for training, which means larger models and longer runs are reachable at all, not just faster.

Full model export, and continue-training from it. Turn on "Export full model" in Training Config and the run hands back a merged 16-bit safetensors checkpoint rather than only the LoRA adapter. That checkpoint drops straight into a serving stack with quantisation left as your decision, and it is what you train again: "Continue from your models" takes a model you already trained as the base for the next run. For client work this is the difference between a retrain that builds on last quarter's model and one that starts over.

How Projects Actually Work

Understanding the project model matters for judging whether Pro fits your workflow, because it is not the multi-tenant structure people often assume.

A project is a workspace inside the Model Studio tab. Each one has its own canvas, its own training-run history, and its own name. What lives outside projects, at the account level, is just as important:

• Datasets live in Data Craft and are reachable from any project
• Model artifacts (LoRA adapters and GGUF exports) live in the Hub and are reachable from any project
• Storage quotas are account-wide, so a project does not get its own slice

So the separation Studio gives you is organisational. One project per model you are actively building keeps canvases from cluttering each other, and naming projects after the client or use case keeps the work legible. Keeping one client's data away from another's is a discipline you apply when you pick which dataset feeds which run, and it is worth being clear-eyed about that if you are handling data under a client confidentiality agreement.

The LoRA adapter model is what makes this affordable at all. Each fine-tuned model is a lightweight adapter, typically 50 to 200 MB, sitting on top of a shared base model. Ten models share base weights and differ only in their adapters, which is how 100 GB goes a lot further than ten full model copies would suggest.

Running Several Trainings at Once

There is no hard concurrency limit on any Ertas plan. Every Action Module on the canvas runs as an independent job, so you can put ten Fine-Tune modules down and press play on all of them.

What actually limits parallelism is credit availability. Starting a run puts its estimated cost on hold, and new runs draw from whatever is not held. If the next run's estimate exceeds your unheld balance, it is blocked at submission. On Pro, 200 credits is what determines how much you can have in flight at once.

Who Pro Is Right For

Consultants and agencies with 3 to 10 active fine-tuning engagements. You are delivering custom models, managing ongoing retraining, and you need the credit budget and storage to keep several models alive at the same time.

Freelance AI consultants juggling concurrent client work. You work solo but rotate between 4 to 6 clients. The credit allowance is what stops you rationing training runs.

Small internal AI teams building models for multiple departments. A marketing model, a sales model and a support model each need their own training data and their own GGUF output. Separate projects give each one a clean workspace.

The ROI Calculation

The question is not whether $50 a month is expensive. It is whether it is expensive relative to what it lets you bill.

Scenario 1: 5 clients at $500/month retainer each

• Monthly revenue: $2,500
• Pro plan cost: $50
• Plan as % of revenue: 2.0%

Scenario 2: 3 clients at $800/month project fee

• Monthly revenue: $2,400
• Pro plan cost: $50
• Plan as % of revenue: 2.1%

Scenario 3: 10 clients at $300/month retainer (model maintenance only)

• Monthly revenue: $3,000
• Pro plan cost: $50
• Plan as % of revenue: 1.7%

The plan pays for itself with a single client paying $70 a month or more. Past one active fine-tuning client, the cost stops being a real input to the decision.

What Each Client Needs (and What Pro Delivers)

The sanity check is worth calling out honestly. When a run finishes, Studio shows three inference samples generated from a small fixed set of prompts. That tells you the model produces coherent, on-format text. It does not tell you the model is good, because the prompts are not yours and nothing is graded. Real evaluation means downloading the model and running your own probe set against it, and the Verifying exports guide walks through that. A built-in evaluation suite is on the roadmap.

Comparing the Cost to DIY Infrastructure

Some agencies weigh building their own fine-tuning setup instead of paying for Pro. The honest comparison:

DIY can make sense for large agencies with dedicated ML engineers. For small agencies and consultants, paying $50 a month for infrastructure that would otherwise cost hundreds in developer time is a straightforward call.

Pro vs Business: When to Upgrade

Business is $100 a month. The upgrade gives you:

• 400 credits/month (vs 200)
• 200 GB model artifact storage (vs 100 GB)
• 10 GB dataset storage (vs 5 GB)
• Higher priority in the GPU queue
• Priority preview access to new features

Upgrade to Business when:

• You are finishing the month with your credit balance at zero
• You are deleting Hub artifacts to make room rather than because you are done with them
• Your datasets have outgrown 5 GB
• Queue position during busy periods is costing you delivery time

Stay on Pro when:

• 200 credits comfortably covers your retraining schedule
• Your exported models and datasets fit inside 100 GB and 5 GB
• Standard priority scheduling is fast enough for your turnaround

Predictable Monthly Pricing

Pro is $50 a month, billed monthly. Cancel anytime.

How to Know You Are Ready for Pro

You are ready for Pro if:

• You are training often enough that Builder's 100 credits run out before the month does
• You are keeping several fine-tuned models alive at once
• You are delivering fine-tuned models as part of a paid service
• You are billing clients for AI work that involves custom models

If you train occasionally and keep one or two models around, Builder at $25 a month is likely enough, and Lite at $10 covers getting past the Free tier's limits. Pro earns its place once training volume, not curiosity, is what sets your monthly usage.

---

Further Reading

• Ertas Builder Plan Review — Is the Builder plan right before upgrading to Pro?
• Agency AI Cost Reduction — How agencies cut costs 99%+ with fine-tuned local models
• Manage Multiple Fine-Tuned Models — Operational workflows for multi-client model management
• White-Label AI Platform for Agencies — When white-labeling becomes important
• Recurring Revenue with Fine-Tuned Models — Building retainer income around model maintenance