The Three-Tier Answer: What OpenAI Actually Charges for Whisper

Ask three sources what Whisper costs and you'll get three different numbers — and all three are technically correct. That's because "Whisper" now refers to three distinct offerings, each priced differently.

The first is the legacy Whisper API (whisper-1), long cited at $0.006 per minute ($0.36/hour). Two independent pricing trackers, including one that compares OpenAI's API against self-hosted transcription, report that rate. But there's a catch: a 2026 technical comparison warns that the standalone whisper-1 rate has "aged off OpenAI's current transcription pricing table" and advises pricing it directly with OpenAI rather than carrying an older number forward. Treat $0.006/min as the historical figure, not a guaranteed current one.

The second tier is realtime transcription. OpenAI's official pricing page lists gpt-realtime-whisper at $0.017 per minute ($1.02/hour) for live transcription — nearly triple the legacy batch rate. Notably, that same analysis flags the realtime endpoint as "not the recommended starting model" for a new integration.

The third tier is the one most people miss: the open-source Whisper model itself is free. But "free" hides real costs. Self-hosting runs an estimated $861–$1,500/month once you count GPU time, storage, monitoring, and DevOps labor — and only breaks even at roughly 2,400 hours of audio per month.

For context, here's how OpenAI's transcription models stack up per minute:

• gpt-realtime-whisper (live): $0.017/min
• gpt-transcribe: $0.0045/min
• gpt-4o-transcribe: $0.006/min
• gpt-4o-mini-transcribe: $0.003/min
• Whisper (legacy API): $0.006/min

Two contradictions are worth knowing before you budget. On free credits, one tracker reports $5 in credits for new OpenAI accounts (roughly 833 minutes, expiring after three months), while another states flatly that OpenAI offers no free tier for the Whisper API — you pay from the first minute. And hosted Whisper rates vary wildly by provider: the same Large v3 weights showed a 2.5× error-rate spread across hosts, with per-minute prices ranging from $0.00067 to $0.006.

Why does this matter for defense work? At $0.36/hour, transcribing the ~7 hours of bodycam an average officer records monthly costs about $2.52 in raw API fees. But that buys only a transcript: no speaker diarization (Whisper doesn't support it; a separate service adds ~$0.005/min), no citations into the video, no redaction, no audit trail. That's the gap a purpose-built workspace like Evidence Copilot fills — its published plans (Solo $39/user/mo with 15 media-processing hours; PD Office $750/attorney/yr) bundle diarized transcription, OCR, and grounded citations rather than metering raw minutes.

Why the Per-Minute Rate Is Not the Real Cost for Evidence Work

The $0.006-per-minute Whisper API rate looks like the cheapest transcription on the market — and it is, right up until you try to use the output in a case. What the base price buys you is a raw wall of text with no speaker labels, no timestamps you can cite, and no way to redact anything before turnover.

Start with diarization. The legacy Whisper model does not support speaker identification — you get one continuous transcript with no speaker labels, which is nearly useless for a two-officer bodycam or a custodial interrogation. Adding a separate diarization service runs roughly $0.005/min, bringing the true full-solution cost to $0.009–$0.016/min, per one pricing analysis of managed transcription stacks.

Then there's the file-size problem. The Whisper API caps requests at 25 MB, and a 28-minute MP3 at 128 kbps is already ~27 MB — so a single shift's bodycam files need chunking before they'll process. An average officer's camera generates about 32 files, 7 hours, and 20GB of video per month, according to TechCrunch coverage of evidence-transcription tools. That's a lot of manual splitting, re-uploading, and stitching transcripts back together.

And even a perfect transcript doesn't get you what evidence work actually requires:

• No citations — nothing ties a quote back to the exact second of video or page of a police report.
• No redaction — faces, PII, and license plates in discovery media stay exposed.
• No audit trail — nothing documents what was reviewed, changed, or produced.
• No OCR — scanned reports and photo exhibits never become searchable text.

The "free" DIY route has its own bill. Self-hosting open-source Whisper runs an estimated $861–$1,500/month in GPU, storage, and DevOps costs, with 40–80 hours of setup and a break-even point around 2,400 hours of audio per month — numbers that only make sense at massive scale, not for a defense caseload. More serious for defenders: OpenAI does not offer a HIPAA BAA for the Whisper API, and academic research on public-defender AI adoption notes that OpenAI discloses ChatGPT records may be used for further training. Running raw discovery audio through a consumer API is a confidentiality problem you'd have to explain to a client.

This is the gap a purpose-built workspace fills. Evidence Copilot's published pricing — Solo $39/user/mo with 15 media-processing hours, Firm $99 with 60, PD Office $750/attorney/yr — bundles diarized transcripts, OCR, timestamped citations, frame-accurate redaction, and an audit trail into every plan, with no feature tiers. At $0.36/hour, 15 hours of raw Whisper transcription costs about $5.40 — but it buys only the transcript. The per-minute rate was never the real cost; the missing workflow was.

The Volume Math: What a Defender's Monthly Bodycam Load Actually Costs

Here's the uncomfortable math: the raw API cost of transcribing a month of bodycam footage is pocket change — a few dollars, not a few hundred. The real cost lives in everything the API doesn't do.

Let's run the numbers. The average officer's body camera records about 32 files, 7 hours, and 20GB of video per month at 720p, according to TechCrunch's coverage of evidence transcription. At the legacy Whisper API rate of $0.006 per minute ($0.36/hour), that 7 hours of monthly bodycam costs roughly $2.52 in raw API fees. Even a heavy month — say 15 hours across multiple officers on one case — runs about $5.40.

Scale that up and the API tiers still look trivial:

• 100 hours/month: $36 on the API, or $66 if you add speaker identification
• 500 hours/month: $180, or $330 with speaker ID
• 2,000 hours/month: $720, or $1,320 with speaker ID
• 5,000 hours/month: $1,800

The self-hosting route looks free until it isn't. Running open-source Whisper on your own infrastructure costs an estimated $861–$1,500/month in GPU, storage, monitoring, and DevOps — and only breaks even against the API at roughly 2,400 hours per month, per one pricing analysis. Most defense offices never approach that volume.

But here's what those dollar figures actually buy: raw transcripts only. No speaker diarization — the legacy Whisper model doesn't support it, and bolting it on adds roughly $0.005/minute through a separate service. No citations that link a quote to the exact second of video. No redaction for discovery turnover. No OCR for scanned reports. No audit trail. And a 25 MB per-request file limit means most bodycam files need chunking before upload.

There's also a confidentiality problem specific to defense work. An academic study of public-defender AI adoption found practitioners cite confidentiality risks and unsatisfactory tool quality as barriers — noting that OpenAI discloses ChatGPT records may be used for further training. Piping discovery audio through a raw consumer API raises exactly the concerns that study documents.

This is the gap a purpose-built workspace fills. Evidence Copilot's published pricing — Solo at $39/user/month with 15 media-processing hours included — bundles diarized transcription, grounded citations, OCR, redaction, and an audit trail into one per-seat rate, rather than metering each capability separately. The raw API is cheaper per minute; the question is what a transcript without speaker labels, timestamps you can cite, or redaction is actually worth in your workflow.

Whisper vs. a Purpose-Built Evidence Workspace: Comparing Published Pricing

At $0.36 per hour, raw Whisper transcription looks almost free — until you tally what it doesn't do. The real comparison isn't per-minute API cost versus subscription price; it's what you get per dollar of discovery review.

Start with the raw numbers. At the legacy Whisper API rate of $0.006 per minute, 15 hours of audio costs about $5.40 and 60 hours about $21.60 — but that buys transcripts only. Speaker diarization isn't included; you'd bolt on a separate service at roughly $0.005 per minute, pushing full-solution costs to $0.009–$0.016 per minute, per one pricing analysis. And a 2026 technical comparison cautions that the standalone whisper-1 rate has aged off OpenAI's current transcription pricing table, so the "cheap" baseline number itself may be stale.

Now put Evidence Copilot's published plans — company product facts, not third-party data — beside that math:

• Solo: $39/user/mo with 15 media-processing hours included
• Firm: $99/user/mo with 60 hours and a priority queue
• PD Office: $750/attorney/yr with 100 hours

The gap between $5.40 and $39 isn't transcription markup. It's everything raw Whisper never delivers: diarized transcripts with word-level timestamps, OCR on scanned police reports and photo evidence, citations that point to the exact second of video, frame-accurate redaction that never touches the original, and an audit trail on every action. Every plan includes the full workspace — there are no feature tiers — and overage is metered and hard-capped at 3x included hours, so a runaway discovery dump can't produce a runaway invoice.

Scale makes the tradeoff concrete. A single officer's body camera generates roughly 32 files, 7 hours, and 20GB of video per month, per TechCrunch's coverage of public-defender evidence tech. Raw Whisper handles that volume for pocket change — then leaves you hand-labeling speakers, hunting timestamps manually, and finding no redaction or audit path at all.

For a defender weighing motion-to-suppress prep or Brady review against a deadline, the per-minute rate is the wrong unit of account. The question is what an hour of reviewed, citable, defensible evidence costs — and that's the number a bundled workspace publishes up front.

How to Decide: A Practical Checklist for Your Discovery Workflow

The per-minute price of Whisper is almost irrelevant to your actual decision. What matters is which tier of work you're doing — and whether raw transcripts are enough, or you need the workflow around them.

When the raw Whisper API makes sense. If you're transcribing one non-confidential recording, running a quick proof of concept, or testing whether automated transcription is even usable for your intake, the math is trivially cheap. A 10-hour proof of concept on the Whisper API costs about $3.60, and the legacy rate of $0.006/minute ($0.36/hour) is among the lowest managed rates available. One caveat: that standalone rate has reportedly aged off OpenAI's current pricing table, so verify the number directly with OpenAI before building anything on it. Also note the 25 MB per-request file limit — a 28-minute MP3 at 128 kbps runs about 27 MB, so longer recordings need chunking.

When self-hosting is justified. The open-source model is free, but the infrastructure isn't. Self-hosting runs $861–$1,500/month in GPU, storage, and DevOps costs, and only breaks even around 2,400 hours of audio per month. If your office processes thousands of hours monthly and has strict data-control requirements — remember, OpenAI offers no HIPAA BAA for the Whisper API, and an academic study of public defenders flags that OpenAI discloses ChatGPT records may be used for training — running your own instance can be the right call. Budget for 40–80 hours of DevOps setup, though.

When a purpose-built workspace wins. For actual case-scale discovery — a dump of bodycam, jail calls, 911 audio, and police reports — raw transcripts leave you doing the hard parts by hand. Whisper's API doesn't include speaker diarization (adding it runs ~$0.005/minute extra, bringing full costs to $0.009–$0.016/min), and it provides no citations, no redaction, and no audit trail. Consider a dedicated workspace like Evidence Copilot when you need:

• Diarized, timestamped transcripts you can cite to the exact second in a motion or cross-exam outline
• Automatic flagging of report-vs-bodycam inconsistencies before you've read every page
• Frame-accurate video and audio redaction for discovery turnover, with originals preserved
• An audit trail your ethics obligations can live with

Evidence Copilot publishes its pricing openly — Solo at $39/user/month with 15 media-processing hours, Firm at $99 with 60 hours, and a PD Office tier at $750/attorney/year — with every plan including the full feature set. At the raw API rate, 15 hours of transcription would cost about $5.40, but that buys transcripts only, with diarization, redaction, and citation work still on you.

The honest framing: Whisper is a component, not a workflow. If you want to see whether a grounded evidence-review workspace fits your practice, you can book a 20-minute walkthrough at evidencecopilot.com/book or start on the Solo plan at evidencecopilot.com/signup. No pressure either way — the DIY stack is a legitimate starting point for small, non-sensitive jobs.