RapidAPI: what would be the best bet?
Research for the owner, 2026-10-01. Numbers scraped from the RapidAPI hub today unless a source says otherwise; the scripts and raw JSON are in C:\Users\Admin\claude\rapidapi_research\.
1. The video, in one paragraph
"How I Turned Code Into A Money Making Machine" (Andy Tries Coding, 5 min, 405k views). Its whole method: build small, single-purpose APIs ("calorie info, currency conversion, analysing headlines"), "pick a niche with demand", keep endpoints clean, price well, list them on RapidAPI and let the marketplace's audience find them. It claims "over $6,000 a month" from "a handful" of APIs, gives no example, no numbers of subscribers, and no API name, and it ends with a link to a paid course. Correction, 2026-10-01 evening, from the NotebookLM grounding check for EP22: the video is "How I Turned Code Into A Money Making Machine" and it points to a video guide with free lessons; "paid course" was wrong. Treat the method as a prompt and the income claim as unverified.
2. How the marketplace works now (facts)
| Fact | Detail | Source |
|---|
| Ownership | RapidAPI was acquired by Nokia in late 2024; the public hub still runs with no documented change to the marketplace mechanics | docs.rapidapi.com, search summary |
| Size | 81,764 APIs listed today. Business 19,607 · Other 12,561 · Data 9,828 · Tools 6,653 · AI/ML 4,432 · Text Analysis 1,020 · Translation 299 | hub search facets, 2026-10-01 |
| Fee | 25% marketplace fee on all payments through the hub, from 15 Nov 2025 (20% before), plus the card-processing fee reported as 2.9% + $0.30 per transaction | Payouts and Finance |
| Payout | PayPal only. A month's charges are paid at the end of the following month (January's money arrives in the first week of March) | same |
| Hosting | You host the API yourself; RapidAPI is a proxy and a billing layer in front of it | Gateway configuration |
| Timeout | 180 seconds is the gateway's default and maximum wait for your server; longer work needs a submit-then-poll design | Hub Listing – Gateway Tab |
| What buyers see | Every listing shows a popularity score (0–10), latency, success rate, the number of subscribers (free tiers included) and the plans | hub listings |
| What sellers earn | Developers who write about it report $150–$800 a month after months of listing; one took almost a year to the first paying customer. The "$2,000–$8,000" figures come from blogs selling guides | community thread, 1xapi blog |
3. What actually gets subscribers (scraped today)
Subscribers include free-plan users, so a count is interest, not revenue. PRO is the first paid plan.
| Listing | What it is | Subscribers | Popularity | PRO / month |
|---|
| ChatGPT 4 · OPEN AI | resold OpenAI access | 21,978 · 10,127 | 9.9 | $5.99 · $4.99 |
| Text-To-Speech | generic TTS, free plan only | 15,659 | 9.8 | — |
| Deep Translate · MyMemory · Rapid Translate | translation | 8,162 · 4,977 · 2,349 | 9.9 · 9.7 · 9.9 | $9 · $299 · $10 |
| Open AI Text To Speech | resold OpenAI TTS | 2,043 | 9.8 | $1 |
| Speech-To-Text AI | generic STT | 1,592 | 9.7 | $10 |
| Whisper: Speech To Text · OPENAI WHISPER · Whisper: From URL | resold Whisper | 634 · 265 · 238 | 0.0 · — · 9.2 | $9.99 · $0 · $9 |
| GerVADER | German sentiment, a language-specific model | 171 | 8.7 | $2 |
| Offensive User Comment Detection | text classification | 125 | — | $9.97 |
| LinkedIn / Facebook / Instagram / TikTok / Amazon scrapers | the hub's front page by relevance | (not read) | 9.9 | — |
| Thai, all 116 results for "thai" plus "thailand" |
| Thai Lottery Result · Thai Lottery · Thai Lottery | lottery draws, three providers | 148 · 127 · 95 | 8.5 · 9.6 · 9.6 | $10 · $2.50 · $0 |
| Oil Thai Price | fuel prices, free only | 125 | 9.3 | — |
| Shopee E-Commerce Data | scraping, popularity 0.1 | 154 | 0.1 | $49 |
| Thai-National-Id (iApp) | Thai ID-card OCR, updated 4 years ago | 13 | 9.3 | $37.50 (ULTRA $150, MEGA $450) |
| Thai Address Parser · Helper · Postal Code · Utility | four address tools | 3 · 4 · 2 · 2 | — | $4.99–$9.99 |
| Thai Macro & Financial Sentiment | the only "thai nlp" result | 1 | 8.5 | $14.99 |
| What we could build, as the hub has it today |
| "thai speech to text", "thai tts", "thai transcription" | no Thai-specific speech API at all; results are generic wrappers | — | — | — |
| Meeting summary APIs (7) | AI Meeting Minutes Suite the largest | 32 · 18 · 5 · 3 · 2 · 1 · 1 | 8.3 (one) | $9–$9.99 |
| Intent classification | one listing | 3 | — | $10 |
Reading. Three kinds of listing get subscribers: a big model resold cheaply (thousands, at $1–$10, a race to the bottom); platform scrapers (the front page, and a terms-of-service problem we do not want); and regional data and language-specific tools (German sentiment 171, Thai lottery 95–148, fuel prices 125). Thai developers are on the hub: three lottery APIs hold 370 subscriptions between them. Nobody offers Thai speech-to-text or Thai text-to-speech there.
4. Our edge, and what is licence-clean to sell
- Measured Thai speech-to-text (EP17, published): on 50 real Thai clips, characters wrong — Whisper large-v3 15.5%, Thonburian whisper-th-large-v3-combined 12.2–12.8%, Thonburian distill 12–13% at a sixth of the time, Typhoon Whisper Turbo 6.7%. Read speech: large-v3 7.1%, Thonburian 5.0–5.4%. Speed on our GPU, one hour of audio: distill 3 min, large-v3 18 min. These are numbers no competitor on the hub publishes.
- Measured summarising (EP19, in progress): transcript → local model → counted "kept / left out / invented". A second endpoint, later.
- Hosting we own: the DGX Spark is right for building and measuring; it is on home internet that moves between networks, so it is wrong for a listing whose latency and success rate are public. Serverless GPU is the fit: Modal bills L4 time at $0.000222 per second and scales to zero, with $30 of free credit a month (modal.com/pricing).
Licences decide the model. Whisper large-v3 is MIT. biodatlab's Thonburian models are Apache 2.0 (the distill checkpoint MIT).
Typhoon Whisper Turbo is MIT by its card but binds users to OpenTyphoon's terms, which forbid using the output "to provide the services that has the similar nature to the Company's models or Services" and "competitive benchmarking" without written consent (
opentyphoon.ai/tac). SCB 10X sells a Thai ASR service. A paid Thai STT API built on Typhoon's model is exactly what those terms describe. Keep it out, or ask SCB 10X in writing first. (The same clause touches EP17's published benchmark; worth a note.)
Thai text-to-speech is blocked for now: every open Thai voice we have checked is non-commercial (F5-TTS, Thonburian TTS, Typhoon2-Audio, Meta's MMS-TTS), Edge TTS is Microsoft's service and not ours to resell, and Google's is paid.
5. The candidates, ranked
| Candidate | Demand on the hub | Competition there | Cost to run | Licence / legal | Verdict |
|---|
| A. Thai speech-to-text, measured accuracy, sync for short files and submit-then-poll for long ones | STT wrappers hold 240–1,600 subscribers each; Thai developers are on the hub; no Thai STT exists | none Thai-specific; generic Whisper resellers at $9–$10 | ≈$0.001–0.002 per audio minute on a Modal L4 with the distill model; large-v3 ≈ $0.007 | clean with Thonburian / Whisper | best bet |
| B. Thai meeting transcript + summary (EP19 pipeline) | meeting-summary listings: 1–32 subscribers | seven generic ones | STT cost plus an LLM | clean | an endpoint on A, not a product |
| C. Thai data feed: gold price, lottery, fuel | proven: 95–148 subscribers | three lottery APIs, free to $10 | near zero (Lambda) | public data; check each site's terms | a cheap way to learn the hub; little money in it |
| D. Thai NLP utilities (word segmentation, romanisation, dates, PII masking) | "thai nlp" has one listing with one subscriber; address tools have 2–4 | low | near zero | PyThaiNLP, Apache 2.0 | no sign of demand |
| E. Thai text-to-speech | TTS is the hub's biggest free category | none Thai-specific | GPU | no commercial open Thai voice | blocked |
| F. Thai ID / document OCR | iApp's listing: 13 subscribers at $37.50–$450 | iApp | GPU | personal data of third parties through our server | no |
| G. Scrapers of platforms | the hub's front page | dozens | proxies, constant breakage | terms of service | no |
6. The bet: "Thai Speech-to-Text — measured"
What it is
POST /transcribe — audio up to 5 minutes, Thai text back with timestamps, in one request (inside the 180-second limit on an L4).POST /jobs + GET /jobs/{id} — longer recordings, up to a few hours, submit and poll.- Options:
model=distill|large-v3, timestamps=true. Later: /summary from EP19.
- The listing shows our measured error rates with their caveats, the way the channel does: on what audio, with which engine, what "wrong" counts.
Unit economics
| Item | Figure | Basis |
|---|
| GPU time, distill model | ≈ 5 min of L4 per audio hour | 3 min per hour on the Spark (EP17), L4 assumed about half as fast |
| Cost per audio minute | ≈ $0.001 | 300 s × $0.000222 |
| OpenAI's Whisper API | $0.006 per minute | the price to undercut; Thai at Whisper-large quality |
| Plans | FREE 60 min/month · PRO $9.99 for 1,000 min · ULTRA $29.99 for 4,000 min · overage $0.008/min | at or below the hub's STT resellers, above cost by 5–8× |
| After the 25% fee | PRO nets $7.49 | plus processing fees |
| Break-even on hosting | about 4 PRO subscribers | Modal's free $30/month covers the first thousands of minutes |
What a month of validation looks like
- Deploy the distill model on Modal (one Python file), with the three endpoints and the RapidAPI proxy-secret check.
- List it with the free plan live and the measured table in the description; no marketing.
- Read what the hub shows after 30 days: subscribers, calls, the success rate, which endpoint is used. That is the demand measurement, in the owner's own method: observed, not argued.
- Only then decide whether to add the summary endpoint, a Thai TTS when a commercial voice exists, or to stop.
Honest expectation. The hub's own numbers say a language-specific tool collects on the order of 100–200 subscribers over a year, mostly free-tier. At a 5–10% paying share that is $50–$150 a month, not $6,000. The upside is that it is the first Thai STT on the marketplace, it reuses work already done and measured, and it costs nothing to find out.
7. What I need from you
- A go-ahead on the bet (A), or a different pick from the table.
- A RapidAPI provider account (sign-up, then PayPal for payouts) and a Modal account (free tier) — both are yours to create; I can write everything up to the moment of signing in.
- Whether to write to SCB 10X about Typhoon. Without their consent the API uses Thonburian and Whisper only.
Sources