Models / LLMs with audio input

LLMs with audio input

Generally-available models that can hear — accept speech or audio as input, from realtime voice to transcription to multimodal understanding. Ranked by published token price.

The short answer

We track 6 generally-available models that can hear — models that accept audio input — from 3 providers. The most affordable by published token price is Gemini 3.1 Flash-Lite (Google) at $0.563 per 1M tokens blended. The largest context window among them is Gemini 3.1 Flash-Lite at 1.0M tokens.

Last verified: 9 Jul 2026 · sourced from official provider documentation

Most affordable audio-capable modelGemini 3.1 Flash-Lite
$0.563/1M blended · $0.25 in / $1.50 out · 1.0M context · Google

Audio pricing varies by provider — some bill per audio-minute rather than per token, and the token figures shown here are the text-token rates. Confirm the audio-specific rate on the provider’s pricing page before budgeting.

The list

6 models that can hear

Generally-available models only, ranked by blended $/1M tokens. Click a header to re-sort, or filter by provider.

Model Status Context Input $/M Output $/M Blended $/M Cutoff
Gemini 3.1 Flash-Lite
Google · text, image, video, audio, pdf
GA1.0M$0.25$1.50$0.5632025-01
GPT-4o mini Transcribe
OpenAI · audio, text
GA16K$1.25$5$2.192024-06-01
Gemini 3.5 Flash
Google · text, image, video, audio, pdf
GA1.0M$1.50$9$3.382025-01
GPT-4o Transcribe
OpenAI · audio, text
GA16K$2.50$10$4.382024-06-01
GPT-Realtime-2
OpenAI · text, audio, image
GA128K$4$24$92024-09-30
Qwen3.5-Omni-Plus
Alibaba · text, image, audio, video
GA

Blended = 0.75 × input + 0.25 × output $/M tokens (a fair single-number cost proxy). Click any header to sort.

FAQ

LLMs with audio input

Which LLMs accept audio input?

We track 6 generally-available models that accept audio input, across 3 providers — from Gemini 3.1 Flash-Lite (Google) at the low end to the flagships from Anthropic, OpenAI, Google and others. The full, filterable list is on this page, ranked by blended price. Prices come from official provider documentation and are refreshed daily.

What is the cheapest audio model?

Gemini 3.1 Flash-Lite (Google) is the cheapest generally-available model we track that can hear, at $0.25 per 1M input tokens and $1.50 per 1M output tokens — a blended $0.563 per 1M. Cheapest on paper is not always best: check the quality, latency and rate limits before committing.

How is audio priced on these models?

It varies. Some providers bill audio input per audio-minute, others tokenize it and bill per token. The prices shown here are the standard text-token rates and are a starting point, not the audio-specific bill — always confirm the audio rate on the provider’s official pricing page before you budget.

Is the cheapest audio-capable model good enough for production?

Not always. Price is only one axis — quality, latency, rate limits and reliability matter just as much, and the cheapest model on paper can cost more once you account for retries or larger prompts. Use this ranking to build a shortlist, then test the top few on your own workload before committing.