Video - Video & Streaming
CDN for Live Courses: Setup & Strategy
Live course streaming in 2026 is a one-to-many video problem with a tight latency budget. Classic HLS with 6-second ...
Per-minute pricing for managed video transcoding services has held in a narrow band. In published 2025 price lists, AWS Elemental MediaConvert's Basic tier charged about $0.015 per HD output minute and Google Cloud Transcoder API about $0.030. For most streaming catalogs, though, the encode is a one-off line item and delivery is the recurring one. This comparison, evaluated November 2026, covers formats, speed, per-minute pricing and API features for the main managed options and self-hosted FFmpeg. It closes with a decision matrix by workload and a worked cost model.
Managed cloud video transcoding is billed per output minute, and every rendition in an adaptive bitrate (ABR) ladder counts separately. At 2025 list prices, a five-rendition ladder (three HD, two SD) cost roughly $3.60 per source hour on AWS Elemental MediaConvert Basic and $7.20 on Google Cloud Transcoder API. Per-title encoding can change the number of renditions.
Cloud video transcoding services differ on six axes, and per-minute price is only one of them. A cheap encoder that cannot emit CMAF with DRM, or that hides ladder control behind presets, costs more in engineering time than it saves on the invoice.
The table lists approximate list prices as published in 2025 vendor documentation. Treat them as planning numbers and confirm current rates before budgeting.
| Service | Approx. price (2025 lists) | Billing model | Output codecs | API features |
|---|---|---|---|---|
| AWS Elemental MediaConvert | Basic: about $0.0075 SD, $0.015 HD per min; Professional tier higher | Per output minute; codec and tier multipliers; reserved queues | H.264, HEVC, AV1, VP9 and others | Job templates, HLS/DASH/CMAF, DRM key exchange, event notifications |
| Google Cloud Transcoder API | About $0.015 SD, $0.030 HD, $0.060 UHD per min | Per output minute by resolution | H.264, HEVC, VP9 | Job templates, HLS/DASH, encryption config, pub/sub notifications |
| Cloudflare Stream | Encoding not billed; about $5 per 1,000 min stored, $1 per 1,000 min delivered | Bundled storage plus delivery | H.264 ABR | Upload API, webhooks, signed playback; little ladder control |
| Mux Video | Basic quality encoding not billed; higher tiers per minute (estimate, check list) | Per minute encoded, stored and delivered | H.264 ABR | Developer-first API, webhooks, per-title ladders, playback analytics |
| Bitmovin Encoding | Quote-based | Per minute with codec multipliers, or contract | H.264, HEVC, VP9, AV1 | Per-title, multi-cloud or own-account deployment, DRM packaging |
| Self-hosted FFmpeg on GPU or CPU | Instance cost only; depends on utilization (estimate) | Compute hours, reserved or spot | Everything FFmpeg links: x264, x265, SVT-AV1, NVENC, QSV, AMF, software VVC | None built in: you own orchestration, retries and packaging |
Managed per-minute services cost roughly a cent or two per HD rendition minute, so ladder size and codec choice move the encode bill more than vendor choice does.
AWS Elemental MediaConvert has the deepest broadcast feature set (captions, HDR, AV1, CMAF with DRM) and reserved queues for steady batches. Limitation: pricing multiplies by tier, codec and pass count, so AV1 or 4K jobs can cost several times the headline HD rate.
Google Cloud Transcoder API has the simplest price sheet: three resolution buckets and nothing else. Limitation: its codec menu is narrower, and teams standardizing on AV1 delivery will need a second encoder.
Cloudflare Stream removes encoding as a line item entirely. Limitation: you trade away ladder and codec control, and you pay per minute delivered rather than per byte.
Mux Video is the fastest path from upload to playback for application developers. Limitation: like Stream, it bundles delivery, which complicates splitting encode and egress across vendors.
Bitmovin targets premium OTT teams that need per-title AV1 and can run it in their own cloud account. Limitation: no public price list, so budgeting requires a sales cycle.
Self-hosted FFmpeg (8.0 shipped in 2025) gives total control and the lowest marginal cost at high utilization. Limitation: you build the job queue, failure handling, packaging and observability yourself.
One correction to older comparisons: Azure Media Services was retired in June 2024. Azure no longer offers a first-party equivalent, so Azure shops run FFmpeg on GPU VMs or use a third-party API.
Desktop tools are not video encoding services, but they still have a place in the pipeline. OBS Studio is for live contribution: it supports NVENC, AMF and Quick Sync AV1, plus WHIP output for WebRTC ingest. It is not a distribution transcoding farm. HandBrake handles offline catalog normalization with SVT-AV1, VideoToolbox, NVENC and QSV. Apple silicon from M3 onward decodes AV1 in hardware but does not encode it, so AV1 work on a Mac runs in software. Shotcut wraps FFmpeg for ingest triage and clean mezzanine files. XMedia Recode and VidBlasterX suit Windows device-profile conversion and small live studios. None of these scale out to automated pipelines.
AV1 is now a practical delivery codec. Hardware AV1 decode ships in current Apple, Intel, AMD and NVIDIA silicon and in most recent flagship mobile SoCs, while hardware AV1 encode is common on datacenter and recent consumer GPUs. VVC (H.266) remains experimental for delivery because device decode support and licensing clarity both lag.
Take a 100-hour VOD catalog encoded to a five-rendition ladder (three HD, two SD). At 2025 list rates, the per-source-hour cost is $3.60 on MediaConvert Basic and $7.20 on Google Cloud Transcoder API. The whole catalog therefore costs about $360 to $720, paid once.
Now assume that catalog drives 100 TB of HLS delivery a month. At the published price lists checked August 2026, Amazon CloudFront pay-as-you-go transfer alone costs $6,965 a month at that volume, or $83,580 over a year. Against that, the encode is under 1 percent of first-year cost.
This changes how the AV1 decision should be framed. Assume (our estimate) a 25 to 30 percent byte reduction, so 100 TB becomes 70 to 75 TB. At CloudFront's average 100 TB rate of about $70 per TB, that saves roughly $1,740 to $2,090 a month. On a CDN billing $4 per TB at the margin, the same reduction saves $100 to $120 a month. Against that saving, you pay for a second ladder, double storage and a fallback path for older devices.
This blog's heuristic: if monthly delivery cost exceeds monthly encode and compute cost by more than three times, renegotiate or replace the egress rate before changing codecs.
BlazingCDN does not transcode. It delivers the packaged HLS, LL-HLS and DASH output that any of the services above produce. As of November 2026, BlazingCDN delivery is priced from $5 per TB down to $2.50 per TB at volume, with one rate for US, Europe and Asia and no request fees. That puts 100 TB a month at $415, and current rates are on the BlazingCDN CDN pricing page.
The volume thresholds below are this blog's planning estimates in output hours per month, not vendor-published breakpoints.
| Workload | Best fit | Why |
|---|---|---|
| Under about 500 output hours a month | Managed API (MediaConvert or Google Transcoder) | Engineering time to run a farm costs more than the per-minute bill |
| App developers who want upload-to-playback in days | Mux or Cloudflare Stream | Encode, storage and player bundled; accept limited ladder control |
| Premium OTT needing per-title AV1 and DRM | MediaConvert Professional or Bitmovin | Broadcast features and CMAF DRM out of the box |
| About 500 to 5,000 steady output hours | FFmpeg on reserved GPU instances, managed API for overflow | High utilization makes compute cheaper than per-minute pricing |
| Above about 20,000 output hours | Owned or colocated GPU, ASIC or VPU hardware | Amortized hardware beats cloud list rates at sustained load |
| Low-latency live contribution | OBS with WHIP to a WebRTC or LL-HLS pipeline | Hardware encode at the source, transcode at the ingest tier |
| One-off catalog normalization | HandBrake or a scripted FFmpeg batch | No recurring service cost for a job that runs once |
The right transcoding choice tracks steady monthly output hours more than feature lists. Most teams should move off per-minute pricing only once utilization stays high.
Pick three clips that stress encoders differently: sports (high motion), animation (flat gradients) and a talking head (low motion). Encode each with SVT-AV1 at a mid preset, a hardware AV1 encoder and x265 medium, plus each managed service you are short-listing. Measure VMAF at the delivered resolution, not the source resolution. Then report two numbers: dollars per output hour, and concurrent real-time streams per dollar-hour for self-hosted options. Note the measurement date beside each result, because both encoder releases and price lists move within a year.
Video decoding turns a compressed bitstream back into raw frames for playback or further processing. Video transcoding chains decode, optional scaling or filtering, and re-encode into a different codec, bitrate or resolution. Every transcoding service decodes first, so hardware decode support (NVDEC, Vulkan, VideoToolbox) directly limits how many concurrent streams one GPU can transcode.
Managed cloud video transcoding typically costs about $0.0075 to $0.06 per output minute at 2025 list prices, depending on resolution and tier. Google Cloud Transcoder API listed roughly $0.030 per HD minute, and AWS Elemental MediaConvert Basic about $0.015. Each ABR rendition is billed separately, so a five-rung ladder multiplies the headline rate.
AWS Elemental MediaConvert, Bitmovin and self-hosted FFmpeg with SVT-AV1 or hardware AV1 encoders all produce AV1 output. Google Cloud Transcoder API focuses on H.264, HEVC and VP9. Bundled platforms such as Cloudflare Stream deliver H.264 ladders. Confirm AV1 pricing separately, since several vendors apply a codec multiplier above H.264 rates.
Self-hosted FFmpeg usually becomes cheaper once GPU instances stay busy most of the day, which this blog estimates at roughly 500 or more steady output hours a month. Below that, idle capacity and the engineering cost of queues, retries and packaging outweigh per-minute savings. Keep a managed API as overflow for spikes.
It depends on your egress rate. AV1 often cuts delivered bytes by an estimated 25 to 30 percent, which is worth thousands of dollars a month at expensive per-TB rates but only around a hundred at low ones. Compare monthly delivery cost with encode cost first; if delivery is more than three times larger, fix the egress rate first.
Export last month's output minutes per rendition and delivered terabytes from your encoder and CDN logs. Price the encode on two managed services from the table and on one reserved GPU instance running FFmpeg. Then price delivery for the same bytes, and again at 25 percent fewer bytes for AV1. The ratio between the two lines tells you whether your next quarter belongs to encoder tuning or to egress. For teams whose packaged HLS and DASH library is the expensive part, the BlazingCDN Video CDN for VOD delivery replicates the library inside the CDN. Its 14-day testing period runs on real production traffic at $5 per TB, with the monthly minimum waived.
Heavy traffic.
Light bill.
The CDN for video and large traffic
Video - Video & Streaming
Live course streaming in 2026 is a one-to-many video problem with a tight latency budget. Classic HLS with 6-second ...
Video - Video & Streaming
To reduce streaming latency, shrink the buffer the player is forced to hold, then make the CDN deliver partial segments ...
Video - Video & Streaming
A live HLS channel with 6-second segments and 100,000 concurrent viewers puts roughly 33,000 requests per second on the ...
Video - Video & Streaming
As of 2026, the difference between a sports stream that feels live and one that feels delayed is mostly settled before ...