Deepgram: Nova-3 vs Microsoft AI: MAI-Transcribe 2
A side-by-side comparison of Deepgram: Nova-3 ($0.02 / minute) and Microsoft AI: MAI-Transcribe 2 ($21.00 / minute) using current USD prices and capabilities from the api14 catalog. Both run through the same OpenAI-compatible API.
Specifications
| Deepgram: Nova-3 | Microsoft AI: MAI-Transcribe 2 | |
|---|---|---|
| Provider | Deepgram | Microsoft |
| Audio | $0.02 / minute | $21.00 / minute |
| Context window | - | - |
| Max output | - | - |
| Input | audio | audio |
| Output | transcription | transcription |
| Tool calling | ||
| Structured output | ||
| Reasoning control | ||
| Added | Jul 15, 2026 | Sep 3, 2026 |
Key differences
- Deepgram: Nova-3 is 100% cheaper on input tokens.
In depth
Deepgram: Nova-3 is a transcription model from Deepgram, and Microsoft AI: MAI-Transcribe 2 is a transcription model from Microsoft. Both sit in the same category on api14, so neither is a general chat or reasoning model that you would use for text generation. You call one or the other when the job is turning audio into a transcript. Deepgram released Nova-3 on 2026-07-15, and Microsoft released MAI-Transcribe 2 on 2026-09-03, so the Microsoft model is the newer of the two.
On capabilities the two are close to identical in shape. Both accept audio as input and return a transcription as output. Neither lists any supported parameters, so you cannot pass sampling controls, temperature or similar knobs through the API for either one. Neither has a published context length, so there is no token window figure to compare, and the qualitative comparison marks the bigger context question as not meaningfully different. The same goes for tool use, structured output and reasoning: none of those are offered by either model, so there is no capability gap to weigh there. The one clear difference is cost. Deepgram: Nova-3 is the cheaper model on input. Output pricing is not distinguished between them in the comparison data, so treat the output side as about the same.
For picking between them, the decision is mostly about price and provider preference rather than features. If your workload is high volume audio transcription and input cost matters most, Deepgram: Nova-3 is the cheaper input option and is the natural default. If you already standardize on Microsoft models elsewhere in your stack, or you want the more recently released transcription model, MAI-Transcribe 2 is a reasonable choice, and the cost difference on input is the main tradeoff you are accepting. Because neither model exposes tunable parameters, you should not expect to shape the output through the API; plan on post-processing the transcript yourself if you need formatting, punctuation cleanup or downstream structuring.
Calling either model through api14 works the same way because the gateway is OpenAI-compatible. You point your client at the api14 base URL, send your audio in the request the way you would for any transcription call, and set the model field. To use Deepgram, set the model id to "deepgram/nova-3". To use Microsoft, set the model id to "microsoft/mai-transcribe-2". Everything else in your request stays the same, so switching providers is a one line change and you can A/B the two on your own audio without rewriting integration code.
There is no strong case in the given facts for one model being better on quality, since no benchmarks or capability differences are listed beyond input price. The only supported reason to pick MAI-Transcribe 2 over Nova-3 is provider alignment or recency, not a documented capability advantage. If input cost is your constraint, Nova-3 remains the better pick.
Switch with one string
Both models use the same endpoint and request shape. Change the model value from "deepgram/nova-3" to "microsoft/mai-transcribe-2" and prompts, streaming and tools keep working.
import fs from "node:fs";
import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://api14.com/api/v1",
apiKey: process.env.API14_API_KEY,
});
const result = await client.audio.transcriptions.create({
model: "microsoft/mai-transcribe-2",
file: fs.createReadStream("audio.mp3"),
});
console.log(result.text);
More comparisons
- AssemblyAI: Universal-3.6 Pro vs Mistral: Voxtral Mini Transcribe 3
- AssemblyAI: Universal-3.6 Pro vs Mistral: Voxtral Small Transcribe 3
- AssemblyAI: Universal-3.6 Pro vs Moondream: Parakeet Ultra
- AssemblyAI: Universal-3.6 Pro vs NVIDIA: Nemotron 3.5 Transcribe
- Mistral: Voxtral Mini Transcribe 3 vs Mistral: Voxtral Small Transcribe 3
- Mistral: Voxtral Mini Transcribe 3 vs Moondream: Parakeet Ultra
- Mistral: Voxtral Mini Transcribe 3 vs NVIDIA: Nemotron 3.5 Transcribe
- Mistral: Voxtral Small Transcribe 3 vs Moondream: Parakeet Ultra