How an AI interview platform launched at $10/mo when LeetCode Premium is $35

Updated October 2026. This post describes the platform as it stood at launch in May 2026, when it was called InterviewAce, Pro cost $10/mo, and everything ran on Cloudflare's and Groq's free tiers. Three things have changed since: the product is now Interview Prep Academy; Pro is $15/mo; and the backend moved to Fly.io, Neon Postgres and Upstash Redis, so the fixed-cost table below no longer matches today's bill. The reasoning about cost per user still holds — the dollar figures are a May 2026 snapshot. Current plans are always on the pricing page.

A reasonable question to ask: if LeetCode Premium charges $35/mo, Exponent $149/mo, Final Round AI $48/mo, and interviewing.io $225 a session — and they're all big, well-funded companies — how did a solo-built tool launch at $10/mo with a free tier?

The honest answer isn't "we're better than them." It's that we had a structurally lower cost base. Below is the full line-by-line breakdown as it stood at launch, so you can decide for yourself whether the math holds.

The fixed costs (May 2026)

At launch the platform ran entirely on Cloudflare's edge platform plus Groq's open-source LLM inference. Both have generous free tiers. Here was every line item:

That's not a typo. At launch the entire infrastructure cost about three and a half dollars a month — so a single $10 Pro subscriber covered roughly three months of it.

The variable cost — per active user

Variable cost is "what extra do I pay for each additional user?" For SaaS this is usually CPU, database, and outbound API calls. For us it's almost entirely LLM inference.

A typical active user (one who shows up daily for prep) generates roughly:

• ~50 LLM calls per month — practice "AI Explain" requests, Study Coach runs, AI Coach chats
• ~3 voice interview sessions per month — each is ~6 minutes of speech + a scoring call

In tokens, that's roughly:

On Groq's free tier this was well within limits. On Groq's paid tier, Llama 3.3 70B is about $0.59 per million input tokens and $0.79 per million output tokens. At those rates:

• 80,000 input × $0.59/M = $0.047
• 55,000 output × $0.79/M = $0.043
• Audio transcription (Whisper paid tier): ~$0.10
• Total marginal cost per active user: ~$0.19/month

At the launch price, a $10 Pro subscriber left roughly $9.81 of gross margin per month after variable cost. A free user cost roughly $0.19/month at worst — almost always $0, because most free users don't use voice or coach features daily.

What that meant at scale (launch pricing)

(The Cloudflare costs do scale eventually — at 10K Pro users some free tiers run out — but the marginal cost stays in the $0.20–$0.40/user/month range. Groq inference scales linearly until you negotiate volume pricing.)

The structural point survives the move off Cloudflare: infrastructure cost grows linearly with users, and revenue grows linearly with users at many times that rate. Moving to a paid host raised the fixed floor, not the per-user slope.

Why competitors charge what they charge

The natural follow-up: "Then how can LeetCode get away with $35/mo? Are they ripping people off?"

No. They have different cost drivers:

1. Hosted on AWS/GCP/Azure at retail rates. The big interview-prep platforms launched 5–10 years ago on the big clouds, and you can't easily migrate a database with millions of users. Their compute + bandwidth costs are a multiple of what a 2026 start-up pays. That's not us being clever; it's us starting in 2026 instead of 2014.

2. Paid LLM APIs (OpenAI / Anthropic). Most AI interview tools use GPT-4 or Claude under the hood. GPT-4 is ~$2.50 per million input tokens, Claude Sonnet is ~$3 per million. Our Groq path costs ~$0.59 — roughly 5x cheaper for similar-quality output on this kind of task. If a competitor's variable cost per user is $1 and ours is $0.19, they need 5x the revenue to break even.

3. Hired engineers, sales teams, paid marketing. LeetCode has 100+ employees. Each one costs $200K+/year fully loaded. They need $20M+ ARR just to cover headcount. We have one person, $0 marketing spend, and rely on organic traffic + Reddit + Product Hunt. Our break-even number of paying users is in the dozens, not the thousands.

4. VC pressure. A company that raised $20M in Series A at a $100M valuation has to justify that valuation. The fastest way is to charge more and grow ARR. A solo founder bootstrapping has no such pressure — we can price for sustainability, not for the next funding round.

None of these are inefficiencies or their fault. They're rational responses to different business models.

So what's the catch?

If it sounds too cheap, here's what you're actually trading off:

• Smaller team — bugs get fixed quickly (often the same day) but you can't expect 24/7 support with named account managers.
• No proprietary human-written library — the practice bank is multiple choice, and how it is built and checked is documented on the methodology page.
• AI scoring, not human scoring — interviewing.io's $225/session gets you a real FAANG engineer for an hour. Pro gets you live AI voice interviews with a scored report. Different tools, different jobs. Most people benefit from doing several AI mocks before booking one human session.
• No fancy features for the sake of features — no mobile app, no gamification beyond a basic streak, no live community chat. We build the things that actually help you do better in an interview.

What this means for indie SaaS in 2026

If you're a developer thinking about building a SaaS, the punchline is: the cost floor for AI-powered tools has collapsed. In 2022 you needed AWS + OpenAI = ~$3/user/month variable cost. In 2026 you can run on open-source LLMs with optimized inference and pay ~$0.20/user/month. That changes pricing math fundamentally:

• You can sustainably charge $10–$15/mo and still have healthy margins
• Free tiers can be genuinely useful (not just "trial then nag")
• A solo founder with 200 paying users at $15/mo = $3,000/mo of revenue = real income

We're not the first to do this — tools like Buttondown, Plausible and Fathom Analytics have been running lean for years. But the AI category specifically has been weirdly stuck on $35–$200/mo pricing because everyone assumed AI was expensive. Open-source LLMs running on optimized inference (Groq's LPU chips) changed the equation.

Try it

If you've read this far you probably want to check whether the actual product is any good, not just the spreadsheet. The free plan doesn't require a credit card:

• Practice mode — 30 free practice questions a month across 30 subjects
• Mock interviews — timed mock interviews, free and not counted against that allowance
• Voice interview — live AI interviews with Zaheen, part of Pro ($15/mo)

Honest invitation: if you sign up and the AI scoring feels unfair or wrong, email me with the session ID. I read every one.