--- title: "Nvidia made batch transcription free. Only the phone call still pays." description: "Nvidia released a free, open-weights 40-language transcription model eight days after Mistral priced batch speech-to-text at fractions of a cent. Per-minute APIs still sell where latency forces streaming, and independent benchmarks keep undercutting the vendors' own clocks there." publisher: "The Inference" section: "Model" published: 2026-10-29T13:20:21.863Z modified: 2026-10-29T13:20:21.863Z canonical: https://theinference.org/article/nvidia-made-batch-transcription-free-only-the-phone-call-still-pays language: en keywords: "Nvidia, Mistral AI, Speech AI, Voice Agents, Open Weights, API Pricing" --- # Nvidia made batch transcription free. Only the phone call still pays. > Nvidia released a free, open-weights 40-language transcription model eight days after Mistral priced batch speech-to-text at fractions of a cent. Per-minute APIs still sell where latency forces streaming, and independent benchmarks keep undercutting the vendors' own clocks there. The price of a transcribed minute just hit zero. On October 28, [Nvidia](https://theinference.org/markets/companies/nvidia) [released Nemotron 3.5 Transcribe, a free, open-weights speech-to-text model covering 40 languages](https://www.unite.ai/nvidia-launches-open-weight-speech-model-and-463-062-hour-audio-dataset/) under the OpenMDW-1.1 license, downloadable from [Hugging Face](https://theinference.org/markets/companies/hugging-face) and licensed for commercial use [4]. By design it does not stream: it transcribes finished recordings, meetings, call archives, clinical dictations, in batch [1][4]. That single omission draws the new front line of the voice-AI market. ## The paid pool is now real-time only For anything recorded, the API bill is optional. Nvidia's own coverage states the consequence: [self-hosting eliminates per-hour API costs that can quickly reach thousands of dollars a month at scale](https://www.androguider.com/2026/10/nvidia-unveils-free-open-40-language.html) [1]. Eight days earlier, on October 20, the Paris-based AI lab [Mistral](https://theinference.org/markets/companies/mistral-ai) had [priced its Voxtral 3 batch transcription at $0.0035 per minute and a full voice-agent pipeline at $0.04 per minute](https://shattered.io/mistral-voxtral-3-speech-ai-pricing-2026/) [2]. Nvidia's free model undercuts even that floor. The mechanism is not charity. Nvidia gives away the weights to drive inference demand for its GPUs, the same playbook [Meta](https://theinference.org/markets/companies/meta) ran with Llama: the hospital or meeting-software startup that adopts the free model still buys Nvidia hardware to run it [1]. Revenue in the batch market therefore shifts from API vendors to the silicon layer, and it falls as customers move workloads on-premises. *Chart: **Batch transcription's price now starts at zero** Bar chart of per-minute speech-to-text pricing: Nvidia's open-weights model at zero, Mistral batch at $0.0035, Mistral realtime at $0.008, Mistral full voice agent at $0.04.* *Published list prices per minute of audio, US dollars, as of October 29, 2026. Nvidia's open-weights model carries no per-minute fee; self-hosting costs are GPU time, not API billing. Sources: AndroGuider, 28 Oct 2026 [1]; Shattered.io, 22 Oct 2026 [2]; Unite.AI, 28 Oct 2026 [4]. [1][2][4]* ## The real-time premium is contested ground Streaming is where the pure-play vendors keep their franchise. Nvidia's write-up concedes that developers needing live voice agents still buy streaming APIs from [ElevenLabs](https://theinference.org/markets/companies/elevenlabs) or voice-API vendor Deepgram [1]. But that premium now rests on quality claims that independent clocks keep discounting. On October 29, speech-AI API vendor [AssemblyAI](https://theinference.org/markets/companies/assemblyai) [shipped its first text-to-speech model at $15 per million characters, undercutting ElevenLabs' $40 and matching Inworld](https://www.marktechpost.com/2026/10/29/assemblyai-tts-1-text-to-speech-api-voice-agents/) [3]. AssemblyAI's own test puts median time to first byte at 47ms [3]. Cekura's independent benchmark, which measures a different metric, median streaming latency, timed AssemblyAI at 159ms against Inworld's 45ms at the same price, ranking AssemblyAI 8th of 19 timed models [3]. Buyers now have both numbers on one page, and the vendor's own clock is the flattering one. *Chart: **Frontier voice pricing halves, from $40 to $15 per million characters** Bar chart of text-to-speech list prices per million characters: AssemblyAI $15, Inworld $15, Cartesia $37.38, ElevenLabs $40.* *List prices per 1 million input characters for hosted streaming text-to-speech, October 29, 2026. Inworld falls to $7 on its Growth plan; Cartesia's range runs to $50 depending on plan credits. Sources: MarkTechPost, 29 Oct 2026 [3]. [3]* *Chart: **At the same $15 price, the independent clock shows a 114ms spread** Bar chart of median streaming latency in milliseconds from Cekura's benchmark: Inworld 45, AssemblyAI 159, ElevenLabs Flash v2.5 177, Cartesia Sonic 3.6 270.* *Median streaming latency as timed by Cekura's independent benchmark of text-to-speech models, milliseconds, lower is better, published October 29, 2026. Inworld and AssemblyAI both list $15 per 1M characters. Source: MarkTechPost, 29 Oct 2026 [3]. [3]* ## Who wins, who bleeds Enterprises and voice-agent builders capture collapsing transcription costs [1][3]. Nvidia converts free weights into paid GPU hours [1]. The squeezed party is the pure-play voice API vendor. If batch workloads migrate to self-hosted open models, the per-minute batch revenue pool shrinks toward zero, and the remaining real-time moat is latency measured in milliseconds by independent benchmarks. One defensible segment remains where cloud API transcription is blocked by data-residency rules: legal, government, financial and healthcare buyers who need audio to stay inside their own walls [1]. But Nvidia's free model is self-hostable too, so even that refuge has a zero-priced competitor standing in it [1][4]. Watch the next independent benchmark of a streaming model: every millisecond it shaves is pricing power the pure-plays cannot defend with a launch-post table [2][3]. ## Takeaway Free open-weights models have zeroed batch speech-to-text pricing, and the revenue left in voice AI sits in real-time streaming, where independent latency benchmarks now set the terms pure-play vendors cannot defend with their own test tables. ## Sources 1. [AndroGuider, Nvidia Unveils Free Open 40-Language Speech-to-Text Transcription AI Model, 28 October 2026](https://www.androguider.com/2026/10/nvidia-unveils-free-open-40-language.html) 2. [Shattered.io, Mistral's Voxtral 3 Debuts at $0.0035 a Minute, 22 October 2026](https://shattered.io/mistral-voxtral-3-speech-ai-pricing-2026/) 3. [MarkTechPost, AssemblyAI Ships AssemblyAI-TTS-1: A $15 per 1M Character Text-to-Speech API for Voice Agents, 29 October 2026](https://www.marktechpost.com/2026/10/29/assemblyai-tts-1-text-to-speech-api-voice-agents/) 4. [Unite.AI, NVIDIA Launches Open-Weight Speech Model and 463,062-Hour Audio Dataset, 28 October 2026](https://www.unite.ai/nvidia-launches-open-weight-speech-model-and-463-062-hour-audio-dataset/) --- Nvidia made batch transcription free. Only the phone call still pays. — The Inference. Canonical: https://theinference.org/article/nvidia-made-batch-transcription-free-only-the-phone-call-still-pays