Modulate

If you’re STILL paying ~$0.25/hr for transcription… You’re probably funding someone else’s margin. Yesterday, we launched Velma Transcribe, our transcription API - delivering best accuracy in the world, at up to 90x lower cost than @DeepgramAI API: modulate.ai/lp/transcripti… (🧵↓)

From meetings to the stands - Modulators took on the Boston Fleet game together 🏒 Grateful for moments that bring the team closer🤞(and for a little Olympic-level inspiration 🇺🇸) @PWHL_Boston #BostonFleet

🚨 This startup might have just killed Deepgram’s model with pricing alone. New STT model: Velma-2 by @modulate_ai $0.03 per audio hour transcribe. Compare that to: - @DeepgramAI Nova-3 - @elevenlabs Scribe v2 - @AssemblyAI Universal-2 …which can cost 10–90x more depending on usage.

Most people are overpaying for transcription because they don't know this tool exists. I tested the new Velma-2 model by @modulate_ai against @DeepgramAI and @elevenlabs. The results? 10–90x lower costs with the same (or better) accuracy on real conversations.

Introducing Velma Transcribe 🏅 A new speech-to-text API that redefines the economics of transcription, delivering high-accuracy transcription at up to 90% lower cost per hour. Built on Modulate’s ELM architecture and built for real-world conversational audio. Benchmarked on AMI Meeting Corpus + Earnings-22. Read more here: bit.ly/4uBHAYX

Our CTO Carter Huffman is speaking at @NVIDIAGTC 2026. He’s joining James Maki from NVIDIA to break down what it actually takes to run real-time voice AI at scale - from GPU infrastructure to low-latency voice transformation in live experiences. If you’re at GTC, come join the session: nvidia.com/gtc/session-ca… 🎤@NVIDIAGTC

Velma Transcribe API is live. Streaming + batch Drop it into your stack in minutes. Docs: modulate.ai/lp/transcripti… If you're building with voice - we'd love to see what you try with it.

Velma-2 transcription runs at: $0.03 per audio hour transcribed That’s up to 10–90x cheaper than most popular APIs. And it still hits 14.9% WER on the AMI benchmark. Cheaper infrastructure → more things developers can build 🤷‍♂️

We launched Velma Transcribe earlier this week. A transcription API that’s 88% cheaper than @DeepgramAI and 92.5% cheaper than @elevenlabs. So naturally, developers started putting it to work immediately. Voice agents. Support tooling. Call analytics. Why? Because the economics changed. (🧵↓)

Don’t trust our benchmarks. Test it yourself. Upload your audio and compare Velma-2 against: - Deepgram Nova-3 - ElevenLabs Scribe v2 - AssemblyAI Universal-2 Same audio. Same test: speechtxt.com/?utm_source=x&…

The reason isn’t magic. It’s architecture. Velma uses Dynamic Ensemble Blocks (DEBs) - multiple specialized models that collaborate and dynamically switch during transcription. Different audio → different model.

@frank94010 @nijokestu

@nijokestu @modulate_ai best value in transcription hands down

Transcription APIs have been quietly bleeding builders dry for too long. High-quality voice-to-text? Sure, if you're ready to pay luxury prices for something that should be infrastructure, not a luxury. @Modulate_ai just launched their Transcription API today and it's a straight-up price disruption. First Test........ modulate.ai/api/speech-to-… speechtxt.com/?utm_source=x&… Competitive accuracy, dramatically lower cost. Jaw dropped. 🧵

@volenpactus @svpino Yep, supported. Dutch and Romanian are both handled. Give it a spin and let us know how it compares once you run your audio through it. Always fun watching people finally get off the $0.25/hr train 👀

@svpino How are smaller languages like Dutch and Romanian with this? Can’t wait to get off of Deepgram et al

Here is the cheapest transcription API in the world. Period. Let's be honest: Transcription APIs are among the most overpriced components of voice AI infrastructure right now. If you are building a voice agent or any sort of conversational AI, a huge chunk of your bill is just transcription. Get more than a few users, and the cost no longer makes sense. The Modulate team launched Velma Transcribe, and I think it's simply the cheapest API I've seen: $0.13 per 1,000 minutes of audio transcription. This is just ridiculous pricing, and every other API I've ever used is way more expensive than that. If you are building voice applications, this is gold.

Fair point. Cheap only matters if quality holds. Velma-2 hits 14.9% WER on the AMI benchmark (messy real meeting audio), which puts it in the same range as leading APIs - just without the $0.25/hr price tag. So the tradeoff isn’t really accuracy vs cost. It’s mostly industry pricing vs what transcription actually costs now. Best way to judge: run your audio through it and see - speechtxt.com

@svpino Cheapest doesn't always mean best value, considering factors like accuracy and support can greatly impact the overall cost of integrating a transcription API. What's the tradeoff here in terms of quality and ease of use. Is it worth the potential savings for most builders.

@SparkifyAI @darshal_ @AssemblyAI @DeepgramAI

@darshal_ @modulate_ai @AssemblyAI @DeepgramAI Three cents per hour sounds almost too good.

I can’t believe nobody caught this. Speech-to-text pricing just quietly got obliterated. A new model from @modulate_ai is charging: $0.03 per hour of audio. For context, that’s: • 5× cheaper than @AssemblyAI • 8.6× cheaper than @DeepgramAI • 13.3× cheaper than @ElevenLabs If you’re building: • voice agents • meeting assistants • call center analytics • conversation AI This isn’t a small optimization. It changes the entire economics of voice products. Worth testing it yourself using this comparison tool: speechtxt.com/?utm_source=x&… Velma Transcribe API by Modulate: modulate.ai/lp/transcripti…

@printinesperad @darshal_ @AssemblyAI @DeepgramAI See it for yourself here: speechtxt.com

@darshal_ @modulate_ai @AssemblyAI @DeepgramAI Would love to see benchmarks against AssemblyAI and Deepgram on real support calls.

@Savage_Manhood @darshal_ @AssemblyAI @DeepgramAI We’re seeing 14.9% WER on the AMI benchmark (real meeting audio). But honestly, the best test is your messiest call - throw it at Velma and see how it holds up.

@darshal_ @modulate_ai @AssemblyAI @DeepgramAI Curious how accuracy compares on messy real-world calls though.

@BrutalMindset @darshal_ @AssemblyAI @DeepgramAI Exactly what we’re seeing. Developers are already putting velma through the wringer - hackathons, side projects, voice agents. When transcription gets this cheap, people stop worrying about cost and start experimenting. We love to see it. 🤓

@darshal_ @modulate_ai @AssemblyAI @DeepgramAI Interesting. Lower transcription costs could unlock way more experimentation with voice agents.

@WealthHatch_ @darshal_ @AssemblyAI @DeepgramAI Throw your messiest Zoom recording at it and see what happens - crosstalk, background noise, bad mics and all: speechtxt.com

@darshal_ @modulate_ai @AssemblyAI @DeepgramAI I’d test this on messy Zoom recordings first before switching.

@DeepgramAI We’re seeing teams switch/upgrade already 👀 Curious what’s keeping others from doing the same? Test it for yourself: speechtxt.com/?utm_source=x&…

Now the fun part - transcription costs: @DeepgramAI: ~$0.25/hr @modulate_ai: ~$0.025/hr ~90% lower cost Up to 10x savings Same messy conversations. For comparable (or better) accuracy 👀

If you’re STILL paying ~$0.25/hr for transcription… You’re probably funding someone else’s margin. Yesterday, we launched Velma Transcribe, our transcription API - delivering best accuracy in the world, at up to 90x lower cost than @DeepgramAI API: modulate.ai/lp/transcripti… (🧵↓)