Google Cloud Speech-to-Text Pricing 2026: Per-Minute Costs, Free Tier and Hidden Charges

Updated on September 19, 2026 by Steven Jones

Google Cloud Speech-to-Text pricing starts at $0.016 per minute for standard V2 recognition, equivalent to $0.96 per audio hour at normal volumes. V2 Dynamic Batch costs $0.003 per minute, or $0.18 per hour, when transcription doesn’t need immediate processing.

If you simply need to upload audio and get a transcript rather than build against a Cloud API, you can use the DIY AI speech-to-text tool. For developers pricing Google specifically, the calculations below use Google’s published rates checked on 18 September 2026 against the Google Cloud Speech-to-Text pricing page.

Google Cloud Speech-to-Text pricing at a glance

Recognition route	Published price	Cost per audio hour	Best fit
V2 Standard	$0.016/min for the first 500,000 monthly minutes	$0.96	Streaming, synchronous and normal-priority transcription
V2 Dynamic Batch	$0.003/min	$0.18	Recorded audio that can wait for lower-priority processing
V1 Standard with data logging	First 60 min/month free, then $0.016/min	$0.96 after allowance	Existing V1 workloads using data logging
V1 Standard without data logging	First 60 min/month free, then $0.024/min	$1.44 after allowance	Existing V1 workloads without data logging
V1 Medical	First 60 min/month free, then $0.078/min	$4.68 after allowance	Medical conversation or dictation models

The $0.016 headline is therefore only the right planning number for a V2 Standard workload below 500,000 minutes per month. Batch method, volume, channel count and API version can all change the bill.

---

---

What 1, 10, 100 and 1,000 hours actually cost

Most teams never reach Google’s first V2 volume threshold. Even 1,000 audio hours is only 60,000 minutes, so all four examples below remain inside the first $0.016 Standard band.

Audio processed	V2 Standard	V2 Dynamic Batch	Practical use
1 hour	$0.96	$0.18	One interview or meeting
10 hours	$9.60	$1.80	Small podcast or research archive
100 hours	$96	$18	Regular transcription workload
1,000 hours	$960	$180	Production transcription workload

For non-urgent recorded audio, the processing class is often a bigger cost decision than the provider. Moving 1,000 hours from V2 Standard to Dynamic Batch cuts the base recognition bill from $960 to $180.

V2 Standard gets cheaper after 500,000 monthly minutes

Google applies lower V2 Standard rates as monthly usage moves through its published volume bands.

Monthly V2 Standard usage	Rate for that usage band
0 to 500,000 minutes	$0.016/min
500,000 to 1,000,000 minutes	$0.010/min
1,000,000 to 2,000,000 minutes	$0.008/min
Above 2,000,000 minutes	$0.004/min

The first threshold is roughly 8,333 audio hours in one month. That means the lower tiers are irrelevant to most smaller applications, even if their annual transcription volume looks substantial.

A 10,000-hour monthly workload contains 600,000 minutes. The first 500,000 minutes cost $8,000, and the next 100,000 cost $1,000, producing a $9,000 V2 Standard recognition bill. Multiplying all 600,000 minutes by $0.016 would incorrectly produce $9,600 because it ignores the second usage band.

Dynamic Batch is the first pricing decision to make

V2 Dynamic Batch costs $0.003 per minute and is available for batch recognition. It is the obvious rate to model first for archives, recorded interviews, podcast back catalogues, post-call processing and other jobs where the transcript does not need to appear while somebody is speaking.

Don’t model Dynamic Batch for live captioning or interactive voice features just because the rate is attractive. Those applications need a processing route designed around prompt results. A useful architecture often splits traffic: urgent speech uses Standard recognition, while completed recordings and backlogs go through Dynamic Batch.

Google now shows 1-year and 3-year savings-plan rates

Google’s current pricing table also displays Gemini Enterprise Flexible Savings Plan rates alongside normal consumption pricing. Do not treat these as the default public price, because eligibility may require opting into the relevant consumption model.

Billing model	V2 Standard first band	V2 Dynamic Batch
Pay as you go	$0.016/min	$0.003/min
1-year Flexible Savings Plan	$0.0144/min	$0.0027/min
3-year Flexible Savings Plan	$0.0128/min	$0.0024/min

Use the pay-as-you-go column for public budgeting unless your organisation has actually committed to a savings plan. Otherwise, a cost model can look 10% or 20% cheaper than the invoice your billing account will receive.

V1 has the free allowance; V2 does not

Google’s published V1 pricing still includes the first 60 minutes per account each month at no charge. V2 Standard starts charging from the first minute at the normal consumption rate.

That makes the phrase “Google Speech-to-Text free tier” slightly misleading in 2026. The recurring allowance belongs to V1. Temporary Google Cloud account credits are separate and should not be used as the long-term unit cost for a production application.

For a new integration, saving at most $0.96 a month through the V1 allowance is a weak reason to choose the older API. Google positions V2 as its latest Cloud Speech-to-Text API, with changes around recognisers, regions, encryption and configuration. Existing V1 applications can still price their current traffic separately.

The hidden cost is usually billing behaviour, not the headline rate

Multiple channels are billed separately

Google bills each recognised audio channel independently. A one-hour two-channel recording can therefore represent 120 billable minutes rather than 60.

At V2 Standard pricing, that changes a one-hour job from $0.96 to $1.92. Dynamic Batch moves from $0.18 to $0.36. This can materially change contact-centre economics, where agent and customer tracks are commonly stored separately.

Do not downmix purely to save pennies if separate channels materially improve the final transcript. The correct comparison is the cost of an accepted transcript, including any manual work required to repair speaker attribution later.

Every successful request is rounded up to the next second

Google measures processed audio in one-second increments. That makes rounding almost irrelevant for hour-long recordings but potentially expensive for thousands of tiny voice clips.

For example, 100,000 clips lasting 1.2 seconds contain 2,000 minutes of actual audio. If each request is rounded to two seconds, the bill is based on about 3,333 minutes instead. At the $0.016 Standard rate, the difference is roughly $32 versus $53.33.

An empty transcript can still be billable

If Google successfully processes a request and returns a response, the audio is billable even if the response contains no transcript. Requests ending in a Google server error are not charged as successfully processed audio.

Your own retry logic therefore deserves attention. A timeout inside your worker does not prove that Google’s processing failed. Blindly sending the same audio again can create another successful processing job. Keep operation IDs and job state rather than treating every uncertain client response as permission to resubmit.

Speech recognition is not the whole Cloud bill

Cloud Storage, compute, logging, queues, and network movement are separate Google Cloud services. Their cost depends on the architecture, so adding an arbitrary percentage for “infrastructure” is less useful than measuring the resources the transcription pipeline actually consumes.

When does another transcription API become cheaper?

At normal V2 Standard volumes, the raw break-even point is simple. A comparable provider charging less than $0.016 per processed minute is cheaper on base recognition cost. For a workload that can use Dynamic Batch, a competitor needs to beat $0.003 per minute to be cheaper on the same narrow measure.

Channel rules can reverse that result. A provider charging more than Google’s headline rate may still cost less for a two-channel call if it includes multiple channels while Google bills each channel independently. Above 500,000 monthly Standard minutes, Google’s declining usage bands also change the comparison.

That is why the recurring practical question isn’t “which API has the smallest number on its pricing page?” A delayed batch recogniser, a real-time speech API and a two-channel call transcription service are different workloads. Our speech-to-text API pricing comparison normalises competing routes when you need a provider-level cost comparison.

Your search, your sources

Make DIY AI a preferred source

See more of our reporting in Google Top Stories, AI Overviews and AI Mode.

Add DIY AI on Google

A better Google Speech-to-Text cost model

For a production estimate, start with channel-minutes rather than file count or storage size. Then apply the processing route and volume tier.

• Measure the audio duration actually submitted for recognition.
• Multiply by the number of channels Google will process.
• Account for one-second rounding at the request level.
• Separate V2 Standard traffic from Dynamic Batch traffic.
• Apply monthly Standard volume tiers rather than one flat rate at scale.
• Track repeat processing caused by retries, overlap or failed application jobs.
• Add the real Cloud Storage, compute and network costs used by your architecture.
• Measure transcript correction time if output quality affects labour cost.

The last measure is easy to underestimate. Saving $0.50 per audio hour is irrelevant if the cheaper route creates several extra minutes of manual correction. Price per accepted transcript is usually a better production metric than price per API response.

Google Speech-to-Text pricing FAQs

How much does one hour of Google Speech-to-Text cost?

One mono audio hour costs $0.96 with V2 Standard recognition at the first usage tier. The same hour costs $0.18 with V2 Dynamic Batch.

Is Google Speech-to-Text free?

V2 does not have a recurring free allowance on Google’s published rate card. V1 includes the first 60 minutes per account each month at no charge. New-account Cloud credits are promotional credits, not a permanent V2 free tier.

Does stereo audio cost twice as much?

If Google recognises two separate audio channels, it bills each channel independently. A 60-minute two-channel recording can therefore create 120 billable minutes.

Is Dynamic Batch cheaper than V2 Standard?

Yes. The published pay-as-you-go rate is $0.003 per minute for Dynamic Batch versus $0.016 per minute for the first V2 Standard volume band. Dynamic Batch is for lower-priority batch processing, not workloads requiring prompt transcription.

What should you budget?

Use $0.96 per mono audio hour as the conservative starting point for a normal V2 Standard workload below 500,000 monthly minutes. Use $0.18 per hour when the same recorded audio can genuinely run through Dynamic Batch.

Then adjust for channels, request rounding, monthly volume tiers and any savings plan your billing account actually uses. If the question is whether Google is worth paying for rather than simply how much it costs, read our Google Speech-to-Text review. Keeping the quality decision separate from the billing calculation makes both pages more useful.