Compare audio AI models and API pricing
Listen, speak and transcribe. One API.
Explore audio models for transcription, speech synthesis, voice applications, and generative audio. Because providers use different billing units, OneInfer ranks prices only within the same comparable unit.
106 audio models are listed today. All of them use one OneInfer API key and pay-as-you-go wallet.
Choose and call a audio model in 4 steps
The list ranks them by price within one billing unit at a time: per minute, per character or per token. Different billing units are never mixed, so the order is fair.
Lowest comparable price: ElevenLabs: Eleven Realtime v1 Mini
Currently the lowest-priced option in the provider price per audio minute group. Different billing units are never mixed.
$0.0800 / audio minute
View pricing and providersAvailable audio models
Comparable models are ordered by their current normalized price. Pricing content refreshed November 9, 2026.
106 models
| Model |
|---|
ElevenLabs: Eleven Realtime v1 MiniLowest priceelevenlabs Eleven Realtime v1 Mini is ElevenLabs' live speech-to-speech model, the faster, lower-cost size, for everyday support and assistant calls. Connect to WS /v1/realtime, stream 8, 16 or 24 kHz PCM (or 8 kHz G.711) microphone audio in and get 24 kHz spoken replies back, using the OpenAI Realtime event format. Speaks 90+ languages with ElevenLabs voices and supports function tools. Billed per connected minute plus the LLM tokens each conversation uses.
|
elevenlabs Eleven Realtime v1 Max is ElevenLabs' live speech-to-speech model, the larger size, for harder questions and multi-step tool use. Connect to WS /v1/realtime, stream 8, 16 or 24 kHz PCM (or 8 kHz G.711) microphone audio in and get 24 kHz spoken replies back, using the OpenAI Realtime event format. Speaks 90+ languages with ElevenLabs voices and supports function tools. Billed per connected minute plus the LLM tokens each conversation uses.
|
 oneinfer Seed Audio 1.0 is ByteDance Seed's unified full-scene audio generation model designed to create speech, dialogue, sound effects, ambience, and other acoustic elements together within a single coherent audio scene. It supports expressive voice generation from text descriptions, reference-audio-conditioned voice generation, multi-character dialogue, multilingual speech, emotion and prosody control, environmental sound design, and fine-grained dialogue timing with approximately 100 ms precision. The model can generate up to two minutes of audio in a single pass and supports continuation for longer content while maintaining character voice consistency. Its architecture combines language-level scene understanding with a diffusion-based acoustic generator operating in a high-fidelity latent audio space.
|
 elevenlabs Eleven v4 Turbo is ElevenLabs' real-time text-to-speech model designed for conversational voice agents, AI assistants, interactive characters, and other latency-sensitive speech applications. It combines the expressive range and voice fidelity of the Eleven v4 family with approximately 100 ms median inference latency and around 150 ms median time to first speech. The model supports more than 90 languages and accents, high-fidelity voice cloning, expressive Audio Tags, pronunciation control, multilingual voice consistency, and streaming speech generation. It is optimized for natural long-running conversations where emotional expression, speaker consistency, responsiveness, and low latency are important.
|
elevenlabs Scribe v2 is ElevenLabs' state-of-the-art speech-to-text model designed for highly accurate transcription of long-form and complex audio across more than 90 languages. It provides precise word-level timestamps, smart speaker diarization for up to 32 speakers, dynamic audio-event tagging, automatic multilingual language detection, keyterm prompting, entity detection, and natural-language transcript editing. The model is optimized for batch transcription, subtitles, captions, meetings, interviews, podcasts, media archives, research, and enterprise transcription workflows, with improved stability across long recordings, pauses, changing speakers, accents, delivery styles, and extended silences.
|
elevenlabs Scribe v2 Medical is ElevenLabs' specialized speech-to-text model fine-tuned for medical and clinical audio. Built on Scribe v2, it improves recognition of medication names, dosages, anatomy, pathology, clinical terminology, and dense medical dictation while maintaining the base model's accuracy on everyday speech. It supports more than 90 languages and includes speaker diarization, precise word-level timestamps, dynamic audio tagging, keyterm prompting, entity detection, PHI-related entity handling, and no-verbatim transcription. The model is designed for clinical documentation, clinician-patient conversations, medical dictation, patient intake, care-coordination calls, and other healthcare transcription workflows where the generated transcript is reviewed by an authorized healthcare professional.
|
 oneinfer Transcribe 1 Pro is Fish Audio's advanced speech-to-text model optimized for interviews, meetings, podcasts, and multi-speaker conversations. It converts audio into text while automatically detecting the spoken language, identifying different speakers with inline speaker markers, and preserving expressive information such as emotions and vocal events including laughter. The model can optionally return timestamped word-level segments, making it suitable for transcription, subtitle generation, meeting intelligence, podcast processing, speaker-aware transcripts, and other structured speech recognition workflows.
|
oneinfer Fish Audio S1 is a multilingual text-to-speech model capable of generating highly expressive speech. It utilizes a fixed vocabulary of preset emotion tags enclosed in parentheses at the beginning of sentences (e.g., sound effects, tone markers) to steer the emotion and delivery of the generated voice.
|
oneinfer Fish Audio S2 Pro is an advanced 4B parameter open-weight text-to-speech model built on a dual-autoregressive architecture. It supports approximately 80 languages and introduces fine-grained, word-level inline tag control, allowing users to embed open-ended natural language instructions inside square brackets directly in the script (e.g., [whispering], [pitch up]). It outputs 44.1 kHz audio and natively supports zero-shot voice cloning from a reference clip.
|
oneinfer Fish Audio S2.1 Pro is a state-of-the-art production-grade neural speech synthesis model optimized for ultra-low latency. With a Time-to-First-Audio (TTFA) as low as ~70-90ms, it is built for live conversational AI and turn-taking dialogue systems. It natively supports 83 languages without separate endpoints, features high-consistency voice cloning from reference audio, and excels at natural prosody for long-form narration.
|
elevenlabs Eleven v4 is ElevenLabs' flagship text-to-speech model designed for highly expressive, emotionally rich, and natural speech generation. Built on a new speech synthesis architecture, it interprets tone, pacing, emotion, speaker identity, and conversational context to produce voice performances with greater realism and control. The model supports more than 90 languages, high-fidelity professional voice cloning, natural multi-speaker dialogue, and inline Audio Tags for controlling emotions, delivery styles, non-verbal expressions, and sound effects. It is optimized for premium content creation, character voiceovers, audiobooks, dubbing, narration, multilingual productions, and other applications where voice quality, expressiveness, and speaker consistency are prioritized.
|
elevenlabs Eleven v3 Conversational is ElevenLabs' real-time text-to-speech model built on Eleven v3 and optimized for natural, expressive, context-aware voice conversations. It is designed for live voice agents and interactive applications where emotional range, conversational awareness, and low latency are important. The model maintains context across dialogue turns and dynamically adapts tone, pacing, and delivery to match the intent and emotional character of a conversation. It supports more than 70 languages, expressive Audio Tags for controlling emotions and non-verbal reactions, streaming speech generation, pronunciation dictionaries, and character-level timing alignment, making it suitable for customer-service agents, AI assistants, phone calls, interactive characters, and other real-time conversational voice applications.
|
elevenlabs Eleven Flash v2.5 is ElevenLabs' ultra-low-latency multilingual text-to-speech model designed for real-time voice agents, conversational AI, interactive applications, telephony, and large-scale speech generation. It generates high-quality natural speech with approximately 75 ms model inference latency while supporting 32 languages and up to 40,000 characters per request. The model balances speech quality, speed, and cost for latency-sensitive production workloads and supports streaming generation, voice cloning, multilingual voice consistency, and multiple audio output formats. It is particularly suited for AI calling agents, customer support systems, chatbots, games, IVR applications, and other real-time experiences where fast time-to-speech is critical.
|
elevenlabs Eleven Flash v2 is ElevenLabs' ultra-low-latency English text-to-speech model designed for real-time conversational applications, voice agents, interactive assistants, telephony, and other latency-sensitive speech workloads. It generates natural speech with under approximately 75 ms model inference latency and supports up to 30,000 characters of text per API request. The model prioritizes speed and responsiveness while maintaining natural vocal performance, making it particularly suitable for AI calling agents, customer-support systems, IVR applications, conversational AI, games, and other real-time voice experiences.
|
elevenlabs Eleven v3 is ElevenLabs' expressive multilingual text-to-speech model designed for natural, emotionally rich voice generation and complex multi-speaker dialogue. It supports more than 70 languages and provides fine-grained control over speech through Audio Tags that can direct emotions, tone, pacing, accents, delivery style, non-verbal reactions, and sound effects. The model can generate realistic conversations with multiple speakers, including interruptions, emotional transitions, and contextual shifts, making it well suited for audiobooks, character performances, games, dubbing, cinematic dialogue, narration, and other creative voice applications where expressiveness and performance quality are prioritized.
|
elevenlabs Eleven Multilingual v2 is ElevenLabs' multilingual text-to-speech model designed for natural, lifelike, and emotionally nuanced speech generation across 29 languages. It maintains consistent voice identity, personality, and speaker characteristics across languages while providing stable, high-quality output for longer-form speech. The model is optimized for voiceovers, audiobooks, e-learning, localization, narration, professional content, and multilingual productions where speech quality and consistency are more important than ultra-low latency.
|
 novita MiMo-V2.6-Flash is Xiaomi's efficiency-balanced open-weight omnimodal Mixture-of-Experts model designed for coding, agentic workflows, computer use, visual understanding, cybersecurity, and long-horizon tasks. It has 309 billion total parameters with 15 billion activated parameters and natively processes text, images, video, and audio within a 1M-token context window. The model uses a sparse MoE architecture with a 681M-parameter MiMo vision encoder, dedicated audio encoders, and a 5-layer multi-token prediction speculative decoder. It is trained using mixed reinforcement learning across coding, general agents, visual tasks, and cybersecurity, targeting strong agentic capability with more efficient inference than the larger MiMo-V2.6-Pro.
|
 oneinfer Gemini 2.5 Flash-Lite is Google's lightweight and cost-efficient reasoning model in the Gemini 2.5 family, optimized for high-throughput, low-latency workloads. It supports multimodal inputs including text, images, audio, video, and PDFs, with optional thinking for applications that need a balance between speed, cost, and reasoning quality.
|
 oneinfer Qwen3.8 Omni Flash is a multimodal Qwen3.8 Flash-class model designed for efficient reasoning, coding, agentic workflows, visual understanding, video understanding, and long-context processing. The underlying Qwen3.8 Flash architecture combines Gated DeltaNet with Qwen Sparse Attention and Mixture-of-Experts layers to improve inference efficiency for long-context and agentic workloads. It supports up to a 1M-token context window and multimodal understanding across text, images, and video.
|
novita MiMo-V2.6-Pro is Xiaomi's flagship open-weight omnimodal Mixture-of-Experts model designed for advanced coding, general agentic workflows, computer use, visual reasoning, cybersecurity, and long-horizon tasks. It has 1.02 trillion total parameters with 42 billion activated parameters and natively processes text, images, video, and audio within a 1M-token context window. The model uses a sparse MoE architecture with a 681M-parameter MiMo vision encoder, dedicated audio encoders, and a 5-layer multi-token prediction speculative decoder. MiMo-V2.6-Pro is trained with large-scale mixed reinforcement learning across coding, general agents, visual tasks, and cybersecurity to improve long-horizon reasoning, tool use, and autonomous task execution.
|
oneinfer Gemini 3.1 Flash-Lite is Google's low-latency, cost-effective multimodal model optimized for high-frequency and high-volume workloads. It supports text, image, video, audio, and PDF inputs and is designed for lightweight reasoning, data extraction, agentic workflows, tool use, and applications where latency and API cost are primary considerations.
|
oneinfer Gemini 2.5 Flash is Google's high-performance multimodal workhorse model designed for reasoning, coding, mathematics, science, document understanding, and agentic applications while maintaining Flash-tier speed and cost efficiency.
|
Gemini 3.5 Flash Lite is Google's high-efficiency multimodal model with upgraded agentic capabilities, optimized for low-cost, high-volume workloads and focused subagents operating inside complex multi-agent systems.
|
Gemini 3 Flash is Google's fast multimodal reasoning model in the Gemini 3 family, designed for advanced multimodal understanding, coding, agentic workflows, and complex reasoning while maintaining Flash-class speed and efficiency. It supports a 1M-token input context window and up to 65,536 output tokens, with native support for text, images, audio, video, and PDFs. The model supports configurable thinking, function calling, computer use, code execution, Google Search and Maps grounding, file search, structured outputs, URL context, context caching, and combined built-in and custom tool use.
|
Gemini 3.6 Flash is a frontier-class Google multimodal Flash model designed for fast reasoning, coding, multimodal understanding, tool use, and scalable agentic applications while retaining Flash-tier latency and efficiency.
|
Gemini 3.7 Flash is a frontier-class Google Flash model focused on high-performance reasoning, coding, multimodal understanding, tool use, and agentic execution with the latency and cost characteristics of the Flash family.
|
Gemini 3.8 Flash is Google's most intelligent Flash model, engineered for long-horizon software engineering, autonomous agents, complex enterprise workflows, advanced reasoning, and multimodal understanding. Building on Gemini 3.7 Flash, it delivers substantial improvements across coding, agentic tasks, and critical multi-step reasoning while retaining Flash-level speed and cost efficiency. It supports a 1M-token context window, configurable thinking levels, function calling, computer use, code execution, search grounding, structured outputs, file search, and URL context.
|
 meta Muse Spark 1.1 is Meta's flagship proprietary multimodal frontier model developed by Meta Superintelligence Labs (MSL). Designed for complex reasoning and agentic tasks, it can actively manage a 1-million-token context window, orchestrate multi-agent systems, and natively process text, image, video, and audio inputs. It excels at long-horizon software development, tool use, and computer-use workflows.
|
meta Muse Spark 1.2 is a proprietary coding-focused frontier model developed by Meta Superintelligence Labs. Co-trained specifically with the Muse Code terminal agent, it features significant improvements in code generation, complex debugging, codebase understanding, and end-to-end developer workflows. It leverages planning, goal conditioning, and context compaction to sustain progress on long-horizon tasks like whole-repository generation and GPU kernel optimization.
|
oneinfer Muse Spark Open is the open-weight model in Meta Superintelligence Labs' Muse Spark family, released in October 2026: a mixture-of-experts with 816B total and 42B active parameters. It reads text, images, audio and video, supports adjustable reasoning and tool calling, and has a 1M-token context window.
|
Gemini 3.5 Flash is Google's high-efficiency multimodal model offering near-Pro-level coding and reasoning at Flash-tier speed and cost. It is optimized for coding, parallel agentic execution, multimodal understanding, and scalable production workloads.
|
oneinfer Gemini 2.5 Pro is Google's advanced Gemini 2.5 reasoning model for complex problem solving, coding, mathematics, scientific analysis, long-context understanding, multimodal reasoning, and sophisticated agentic workflows.
|
Gemini 4 Argon API on OneInfer: Google's flagship reasoning model with a 1M-token context. $2/1M input, $10/1M output tokens until Nov 30, 2026.
|
Gemini 3.1 Pro is Google's advanced intelligence model for complex reasoning, problem solving, software engineering, multimodal analysis, long-context tasks, and powerful agentic and vibe-coding workflows.
|
deepinfra Kokoro is a fast, low-cost 82M open text-to-speech model with 54 preset voices in several languages.
|
meta Muse Wonder 1.0 is Meta's frontier multimodal reasoning and agentic AI model developed by Meta Superintelligence Labs. It is designed for complex software engineering, long-running autonomous workflows, professional knowledge work, and advanced computer use. The model supports a 1,048,576-token context window, text, image, audio, video and PDF understanding, tool calling, structured outputs, search grounding, and configurable reasoning effort from minimal to max. It introduces Active Memory for retaining relevant information across conversations and Active Focus for sustaining multi-step tasks, recovering from errors, delegating to sub-agents and verifying results. Muse Wonder is optimized for coding, financial analysis, legal research, healthcare reasoning, browser automation, desktop interaction and agent orchestration.
|
 oneinfer MAI-Transcribe-2 is Microsoft's high-accuracy multilingual speech-to-text model designed for fast, production-scale transcription across 60 languages and challenging real-world audio conditions. It provides automatic language identification, speaker diarization, word-level and segment-level timestamps, keyword biasing for domain-specific terminology, code-switching support, and configurable clean or verbatim transcription styles. The model is optimized for meetings, call-center conversations, clinical notes, legal documentation, captions, accessibility, media transcription, content workflows, and voice-agent applications. It combines high transcription accuracy with extremely high batch throughput, processing approximately one hour of audio in around 10 seconds under Microsoft's benchmark conditions.
|
deepinfra Chatterbox Turbo is Resemble AI's efficient 350M open text-to-speech model.
|
deepinfra Chatterbox Multilingual is Resemble AI's open text-to-speech model for 23 languages.
|
deepinfra A 0.6B multilingual text-to-speech model for 11 languages.
|
deepinfra Orpheus is a Llama-based speech model for expressive, human-like speech, with seven preset voices. Up to 300 characters a request.
|
deepinfra CSM is Sesame's conversational speech model, for short, natural-sounding lines of up to 200 characters.
|
oneinfer MAI-Voice-2-Flash is Microsoft's ultra-fast, low-latency, high-fidelity expressive text-to-speech model optimized for real-time conversational voice applications. It provides natural and expressive speech generation with granular emotion control, multilingual synthesis, zero-shot voice prompting from short reference audio, and high speaker fidelity while prioritizing responsiveness. The model is designed for latency-sensitive applications including call-center agents, AI voice assistants, IVR systems, conversational agents, customer support, and other interactive speech experiences where rapid speech generation is critical.
|
oneinfer MAI-Voice-2.1-Flash is Microsoft's ultra-fast, low-latency multilingual text-to-speech model optimized for high-volume real-time voice applications. It produces high-fidelity, natural and emotionally expressive speech across 23 supported languages while prioritizing responsiveness and serving efficiency. The model supports granular emotion and speaking-style control, multilingual synthesis, cross-language speakers, and zero-shot voice prompting or gated instant voice cloning from short consented reference audio. It is designed for latency-sensitive applications including AI voice agents, call-center agents, voice assistants, IVR systems and other real-time conversational experiences.
|
oneinfer Qwen-Audio-3.0-TTS-Plus is Qwen's high-quality text-to-speech model designed for natural, expressive, and controllable speech synthesis. The Plus variant prioritizes synthesis quality, naturalness, timbre fidelity, and detailed expressiveness for professional speech-generation workloads. It supports multilingual speech synthesis, Chinese dialects, voice cloning, natural-language instruction control, and fine-grained control over emotion, tone, character, speaking rate, volume, and synthesis style. The model is also designed to remain robust when working with reference audio containing noise or reverberation, making it suitable for content creation, audiobooks, film and video dubbing, brand voices, virtual characters, and premium speech services.
|
oneinfer MAI-Voice-2 is Microsoft's high-fidelity multilingual text-to-speech model designed for natural, expressive, and emotionally controllable speech generation. It supports granular emotion and speaking-style control, stable speaker identity across long-form content, multilingual synthesis, and code-switching for supported language pairs. The model also provides zero-shot voice matching from a short consented reference recording without requiring fine-tuning, making it suitable for audiobooks, podcasts, narration, accessibility, branded voices, customer support, voice-over, and other applications where speech quality and speaker consistency are prioritized.
|
oneinfer MAI-Voice-2.1 is Microsoft's highest-fidelity and most expressive text-to-speech model in the MAI-Voice family, designed to generate natural, emotionally rich and studio-quality speech across 23 languages. It provides human-like intonation, rhythm and prosody, granular emotion and speaking-style control, stable speaker identity across long-form generations, multilingual synthesis, and zero-shot voice prompting from short consented reference audio without additional fine-tuning. The model is optimized for applications where speech fidelity and expressiveness are more important than minimum latency, including audiobooks, podcasts, narration, educational content, media production, branded voices and professional voice-over workflows.
|
Gemini 3.8 Flash-Lite TTS is Google's fast and cost-efficient text-to-speech model optimized for high-throughput production, real-time voice-agent cascades, read-aloud applications, voice replication, and everyday speech generation. It combines low latency with expressive and natural speech while using the same API schema and structured prompting format as Gemini 3.8 Flash TTS. The model supports 101 languages, single-speaker and multi-speaker generation, prebuilt and custom voices, voice design, voice replication, inline vocal events, and structured control over speaking style, accent, pace, and tone.
|
 deepinfra A low-latency 0.6B speech recognition model from NVIDIA for 40+ languages.
|
 deepinfra Voxtral Small is Mistral's 24B audio model for accurate transcription.
|
openai gpt-realtime-mini is OpenAI's low-cost live speech-to-speech model. Connect to WS /v1/realtime, stream 24 kHz PCM microphone audio in and get spoken replies back with low latency, using the OpenAI Realtime event format. Supports function tools and input transcription with gpt-4o-mini-transcribe. Billed by the text and audio tokens each reply uses, so silence costs nothing.
|
 deepinfra A faster, pruned Whisper Large v3 for low-cost multilingual transcription.
|
deepinfra The compact Qwen3-ASR model: speech recognition and language identification in 30 languages and 22 Chinese dialects.
|
deepinfra Whisper Large v3 is OpenAI's open speech recognition model for multilingual transcription.
|
deepinfra The flagship Qwen3-ASR model: accurate speech recognition in 30 languages and 22 Chinese dialects.
|
 sarvam Bulbul v4 Flash is Sarvam AI's low-latency text-to-speech model for Indian languages, built for voice agents, IVR and other real-time speech. It has 224 voice personas, each one speaker in one language and style, across Hindi, English, Hindi-English code-mixed, Bengali, Gujarati, Kannada, Marathi, Punjabi, Tamil, Telugu and Assamese, with pace, pitch and loudness controls. A request takes up to 2,500 characters.
|
deepinfra Qwen3-TTS is Alibaba Qwen's expressive, low-latency text-to-speech model for 10 languages.
|
deepinfra Qwen3-TTS VoiceDesign speaks in a voice you describe: send the description as the voice_id.
|
deepinfra Higgs Audio V2.5 is a natural-sounding text-to-speech model with clear, steady delivery for short and long texts.
|
deepinfra Fast multilingual text-to-speech from Inworld AI, with 130+ preset voices in 15 languages, built for low latency.
|
sarvam Bulbul v3 is a production-grade text-to-speech model optimized for 11 Indian languages. It features native support for code-mixed speech, professional voice cloning, and industry-leading stability in telephony environments (8 kHz). It automatically infers prosody, emphasis, and emotional tone.
|
deepinfra Low-latency text-to-speech from Inworld AI that follows tone and emotion directions in the text, such as "[be happy] Hello!".
|
Gemini 3.8 Flash TTS is Google's flagship creative text-to-speech model designed for studio-grade voice generation, expressive acting, long-form narration, complex multi-speaker dialogue, regional accents, and precise pronunciation. It provides high acoustic fidelity, natural cadence, strong adherence to speaking-style instructions, and stable voice identity across extended conversations and narrations. The model supports 130 languages, single-speaker and multi-speaker generation, prebuilt and custom voices, voice design, voice replication, regional dialects, IPA pronunciation overrides, inline vocal events, and fine-grained control over style, accent, pace, and tone.
|
deepinfra Inworld AI's higher-quality multilingual text-to-speech, with 130+ preset voices in 15 languages.
|
 minimax MiniMax Speech 2.8 Turbo is the speed-optimized variant of MiniMax's flagship text-to-speech model. Delivering sub-250 millisecond latency, it provides fast, natural, and expressive voice synthesis ideal for real-time applications like conversational AI, voice agents, and interactive experiences. It retains the core advanced capabilities of the HD tier—including 40+ language support, rapid 10-second voice cloning, emotion control, and native sound tags—while prioritizing high-throughput generation and cost-efficiency.
|
minimax MiniMax Speech 2.8 HD is the flagship studio-grade text-to-speech model from MiniMax. Built on an autoregressive Transformer architecture with a Flow-VAE decoder, it delivers highly expressive, broadcast-ready voice synthesis. It features native sound tags for realistic paralinguistic sounds (such as laughs, sighs, and breaths), robust emotion control, and high-fidelity voice cloning from just 10 seconds of audio. Supporting over 40 languages, it is optimized for professional audiobook narration, podcast production, and high-end video voiceovers.
|
deepinfra Voxtral Mini is Mistral's compact audio model for transcription.
|
sarvam Saaras v3 is a state-of-the-art speech recognition and translation model. It natively supports streaming for real-time applications and features specialized modes for transcription, direct-to-English translation, and transliteration. It is specifically tuned for noisy environments and complex code-mixed (e.g., Hindi-English) speech.
|
sarvam Saarika v2.5 is Sarvam AI's legacy speech recognition model designed for Indian languages and accents. It transcribes audio in the same language spoken, excelling in multi-speaker conversations, telephony audio (8kHz), and code-mixed speech. Supports 11 languages (Hindi, Bengali, Gujarati, Kannada, Malayalam, Marathi, Odia, Punjabi, Tamil, Telugu, English) with automatic language detection and speaker diarization. Achieves 4.96% CER and 18.32% WER on VISTAAR benchmark.
|
oneinfer MAI-Transcribe-1.5 is Microsoft's high-accuracy multilingual speech-to-text model designed for fast production-scale transcription across 43 languages. It combines low word error rates with very high transcription throughput and is optimized for meetings, calls, media, enterprise transcription, subtitles, voice applications, and noisy real-world audio. The model supports automatic multilingual transcription and keyword biasing for domain-specific terminology such as names, products, medical terms, and internal acronyms. Microsoft reports that keyword biasing can reduce word error rate by up to 30% on FLEURS. The model is optimized for batch transcription and can process an hour of audio in under 15 seconds under Microsoft's high-throughput measurements.
|
 grok Grok TTS 1.0 is xAI's text-to-speech model designed for expressive, multilingual, and low-latency speech synthesis. It supports natural-sounding speech generation with built-in and custom voices, fine-grained delivery control through speech tags, pronunciation control using IPA and phrase replacements, adjustable speaking speed, and character-level timestamps. The model supports both batch and real-time streaming synthesis and can produce high-fidelity audio as well as telephony-optimized output, making it suitable for voice agents, narration, conversational applications, accessibility, call-center systems, and other production speech-generation workloads.
|
grok Grok Voice 3.0 is xAI's live speech-to-speech model, released in October 2026. Connect to WS /v1/realtime, stream microphone audio in and get spoken replies back with low latency, using the OpenAI Realtime event format. Supports function tools and sessions of up to 120 minutes. Billed per connected minute, plus a small fee per text message.
|
grok Grok Voice Transcribe 2.0 is xAI's speech-to-text model. It transcribes 38 languages with word-level timestamps, from an uploaded file or a URL.
|
deepinfra NVIDIA's open 30B-A4B speech model for offline transcription in 40 languages, for recordings up to an hour.
|
openai GPT Transcribe is OpenAI's speech-to-text model for recorded audio. It takes prompt context and language hints for domain terms and multilingual audio, and is billed by the length of the audio.
|
deepinfra ACE-Step makes a 30-second to 8-minute song from a prompt, with lyrics or as an instrumental.
|
fal Wan 3.0 Prime is Alibaba's accelerated multimodal video generation model designed for high-quality video creation with significantly faster generation speed. It supports text-to-video, image-to-video with first-frame or first-and-last-frame conditioning, and reference-to-video generation using multimodal references including text, images, video, and audio. The model supports native audio generation, enhanced reasoning for complex prompts, multiple aspect ratios, resolutions up to 1080p, and video durations of up to 30 seconds.
|
fal Wan 3.0 Video is Alibaba's high-speed AI video generation model from the Wan 3.0 family. It delivers the core capabilities of the standard Wan 3.0 Video model with significantly faster end-to-end generation speeds, making it ideal for high-volume API workflows and production. It can generate up to 30 seconds of synchronized audio and video in a single pass (at 480p, 720p, or 1080p resolution). The model supports extensive multimodal references, natively accepting text, images, audio, video.
|
fal OmniHuman 1.5 from ByteDance turns one portrait image and an audio track into a talking, gesturing video, with an optional prompt to guide the performance. Audio up to 30 s at 1080p or 60 s at 720p. Billed per second of output video.
|
fal OmniHuman 2.0 from ByteDance turns one image and an audio track of 4 seconds to 5 minutes into a 720p talking, gesturing video, for presenters, explainers, marketing and music videos. An optional prompt guides the scene, movement and camera. Billed per second of output video.
|
fal Aurora from Creatify turns one image and an audio track into a speaking or singing avatar video at 480p or 720p, with an optional prompt. Billed per second of output video, rounded up.
|
elevenlabs Eleven Music v1 is ElevenLabs' first-generation studio-grade music generation model designed to create complete musical tracks from natural-language prompts and structured composition plans. It supports a wide range of genres and musical styles, including instrumental music and songs with generated vocals and lyrics. Users can control track structure, style, section duration, lyrics, instrumentation, and whether the result should be instrumental. The model can generate tracks from 3 seconds up to 10 minutes and supports multilingual vocal generation. It is suited for background music, game soundtracks, podcasts, advertising, social content, songwriting, and other music-production workflows.
|
elevenlabs Eleven Music v2 is ElevenLabs' second-generation studio-grade music generation model designed to create complete songs, instrumentals, soundtracks, and complex musical compositions from natural-language prompts, structured composition plans, and audio references. It improves on Music v1 with stronger prompt understanding and adherence, more natural multilingual vocals, improved instrumentation and arrangement, and support for complex vocal delivery such as fast rap and dense lyrical patterns. Music v2 supports long-form section-by-section composition, mid-track genre transitions, Audio Reference for matching the style and feel of an uploaded track, music inpainting, embedded sound effects, generated or user-defined lyrics, and detailed control over musical structure. It is designed for songwriting, background music, film and video soundtracks, advertising, games, podcasts, social content, and professional music-production workflows.
|
elevenlabs Eleven Music v2.5 is ElevenLabs' most advanced music generation model, designed for creating studio-grade songs, instrumentals, soundtracks, and complex musical compositions from natural-language prompts, structured composition plans, and audio references. It builds on Music v2 with improved audio quality, stronger prompt adherence, richer melodies, deeper arrangements, more realistic instrumentation, and better coherence across complete tracks. The model supports multilingual vocals, generated or user-provided lyrics, complex vocal performances, Audio Reference for guiding style and production characteristics, section-by-section composition, music inpainting, and detailed control over genre, mood, instrumentation, structure, and arrangement. It is designed for songwriting, film and video soundtracks, advertising, games, podcasts, social media, background music, and professional music-production workflows.
|
Lyria 3 Clip is Google's music generation model optimized for creating short, high-quality musical clips, loops, previews, and soundtrack concepts. It generates 30-second stereo music from text prompts or image inputs and supports instrumental music, vocal generation, generated or user-provided lyrics, and detailed musical controls such as BPM and intensity. The model is designed for rapid music ideation and short-form creative workflows, producing MP3 audio together with lyrics and song-structure information. Lyria 3 uses a latent diffusion architecture operating over temporal audio latents and includes audio watermarking and content-safety filtering for generated music.
|
Lyria 3.5 is Google's advanced music generation model designed for creating high-quality, full-length songs from text prompts and image inputs. It produces structurally coherent music with multiple verses, choruses, bridges, vocals, lyrics, and complete instrumental arrangements, with controllable track duration up to approximately three minutes. The model supports detailed creative direction over genre, instrumentation, tempo, dynamics, vocal style, acoustic characteristics, and song structure, while also allowing users to provide their own lyrics or request instrumental-only music. Lyria 3.5 generates professional-grade 44.1 kHz stereo audio and is designed for songwriting, soundtrack creation, advertising, content production, music prototyping, and other professional creative workflows.
|
 fal Kling AI Avatar v2 Pro turns one image of a person, animal, cartoon or stylized character and an audio track into a higher-quality talking video, with an optional prompt. Billed per second of output video.
|
fal Kling AI Avatar v2 Standard turns one image of a person, animal, cartoon or stylized character and an audio track into a talking video, with an optional prompt. Billed per second of output video.
|
fal Kling Lip Sync drives the speaker in a 2-10 second 720p or 1080p video with an audio track of up to 60 seconds. Billed per started 5 seconds of video.
|
 lightricks LTX-2.3 Fast is a speed-optimized variant of Lightricks' LTX-2.3 video generation model. It is designed for high-speed, cost-effective generation of synchronized audio and video in a single pass. It supports up to 20-second clips, native 9:16 portrait and 16:9 aspect ratios, and resolutions up to 4K. Delivering significantly faster inference than the Pro version, it is ideal for rapid prototyping, batch generation, and short-form social media content.
|
lightricks LTX-2.3 is a DiT-based audio-video foundation model designed to generate synchronized video and audio within a single model. It brings together the core building blocks of modern video generation, with open weights and a focus on practical, local execution.
|
lightricks LTX-2.3 Pro Audio-to-Video makes a video timed to an audio track: dialogue, music or ambient sound. The audio sets the length (2 to 20 seconds at 720p and 1080p, up to 10 at 1440p and 4K) and is the video's sound track. Describe the video in the prompt, give it a start frame (and an end frame), or both. Send the audio in audio_urls. Billed per second of audio.
|
lightricks LTX-2.5 Fast is an accelerated endpoint for Lightricks' LTX-2.5 video generation model. Designed for high-speed text-to-video and image-to-video workflows, it trades a degree of rendering fidelity for significantly faster inference speeds. It supports up to 20-second clips, 4K resolution output, and native synchronized audio generation, making it ideal for rapid prototyping and high-throughput production.
|
lightricks LTX-2.5 Fast Audio-to-Video makes a video timed to an audio track: dialogue, music or ambient sound. The audio sets the length (2 to 20 seconds at 720p and 1080p, up to 10 at 1440p and 4K) and is the video's sound track. Describe the video in the prompt, give it a start frame (and an end frame), or both. Send the audio in audio_urls. Billed per second of audio.
|
lightricks LTX-2.5 Pro is the quality-optimized API endpoint of Lightricks' LTX-2.5 audio-video model. It utilizes Diffusion Fidelity Rendering to allocate maximum compute to complex scenes, ensuring high pixel quality, sharp faces, legible text, and seamless multi-shot continuity. It is designed for professional text-to-video and image-to-video workflows with synchronized native audio.
|
lightricks LTX-2.5 Pro Audio-to-Video makes a video timed to an audio track: dialogue, music or ambient sound. The audio sets the length (2 to 10 seconds) and is the video's sound track. Describe the video in the prompt, give it a start frame (and an end frame), or both. Send the audio in audio_urls. Billed per second of audio.
|
fal MiniMax H3 is a general-purpose, omni-modal generative system that supports unified understanding of multimodal contexts composed of text, images, video, and audio. It can generate video with native stereo audio at resolutions up to 2K and durations of up to 15 seconds. It is released with open weights under the MiniMax H3 Community License.
|
deepinfra P-Video is Pruna AI's fast video model: 1 to 10-second clips at 720p or 1080p from a prompt, a start image or an audio track.
|
deepinfra P-Video Avatar makes a portrait speak: the input audio, or the prompt read in a generated voice.
|
fal Lipsync 2 Pro from Sync Labs matches the speaker's lips in an existing video to a new audio track, with detail-preserving face rendering. Billed per second of output video.
|
fal Lipsync 3 is Sync Labs' current lip-sync model. It re-animates the speaker's mouth in an existing video to match a new audio track, keeping the rest of the frame, at up to 4K. Billed per second of output video.
|
fal Fabric 1.0 from VEED animates one image into a talking video from an audio track, at 480p or 720p. Billed per second of output video.
|
fal Fabric 1.0 Fast from VEED animates one image into a talking video from an audio track, faster than Fabric 1.0, at 480p or 720p. Billed per second of output video.
|
 fal Vidu Q4 (preview) keeps the people and objects of up to 12 reference images, and the voices of up to 3 audio clips, consistent across a 3-16 second video at up to 4K, with optional sound. Refer to them in the prompt as [@reference_image_1] and [reference_audio_1]. Billed per second of requested video.
|
deepinfra Wan2.7 animates a start frame (and optional end frame), or continues a clip, for 2 to 15 seconds at 720p or 1080p, with optional audio driving the motion.
|
grok Grok Imagine Video 1.5 is xAI's advanced video model built on the Aurora-2 engine. Its standout feature is native one-pass audio generation, which produces synchronized dialogue (lip-sync), sound effects, and background music simultaneously with the video. Supporting resolutions up to 1080p and durations up to 15 seconds, it allows for complex workflows including text-to-video, image-to-video, video extension, and multi-image reference guidance to maintain consistent styles and characters.
|
How to choose
- Voice agents
- Transcription
- Speech synthesis
- Audio generation
Pricing methodology
OneInfer compares only positive prices with the same billing unit. Per-minute, per-character, per-token, per-image, per-video, and per-second rates remain separate. Prices can change, so the current model page and console remain the source of truth.
Frequently asked questions
What is the cheapest audio model on OneInfer?
The lowest-priced model is calculated from the current catalog within each comparable billing unit. Per-minute, per-character, and token-based prices are not mixed.
Why are some audio models not ranked together?
Audio providers charge by different units. A transcription model billed per minute cannot be honestly ranked against a speech model billed per character without a common workload assumption.