✦  Whisper speech-to-text is live in the model catalog — run it on the fleet today

The world's devices,
one compute fabric

Run inference and training on Bright Data's global network of consumer devices. Submit a batch, a real-time request, or a training job — the platform handles placement, pipelines, and results at fleet scale.

Onboarding is personal — we provision your account and first project with you.
Batch, real-time & training one API for all three
Compute tiers to fit the job everyday CPUs to CUDA GPUs
Server-side pipelines OCR, speech, audio & PDF built in
Per-job billing pay for completed work, nothing else
Powered by Bright Data's consumer-device network
Windows SDK Python SDK Node.js SDK CPU · DirectML · CUDA Devices across the globe

Batch inference at fleet scale

Point the API at your data — inline items, a manifest, a storage folder, or a zip — and the scheduler fans it out across the fleet. Runs survive restarts, results are durable, and progress is visible from the first item.

  • Manifest, folder-scan and staged zip/drive ingestion
  • Restart-safe runs — a deploy loses nothing
  • Webhooks on completion, durable results storage
# one call, any input shape POST /api/v1/inference/batch { "model_id": "catalog-siglip", "data_source": { "zip_url": "…/images.zip", "expected_count": 5000 }, "options": { "labels": ["a cat", "a dog"] } } # → 202 · poll or webhook · results at /results/:jobId
Catalog modelsvision · speech · text
Your modelsversioned uploads
Open vocabularyyour labels, no training

Models ready to run — or bring your own

A curated catalog of permissively-licensed models runs out of the box: image classification, open-vocabulary labeling, Whisper speech-to-text. Or upload your own — versioned, account-scoped, one model behind many projects.

  • Open-vocabulary classification with your own label set
  • Immutable model versions with a pinned production default
  • Per-project adapters, schemas and label overrides

Pipelines that meet your data where it is

Real-world data isn't tensors. The platform decodes, extracts, segments and normalizes server-side — videos become audio, PDFs become pages, speech becomes utterances — before the fleet ever sees a task.

  • Video → audio extraction, VAD utterance segmentation
  • PDF and document-type classification, built-in OCR
  • Explicit processor pipelines, validated before dispatch
# a video is one item — the platform does the rest video.mp4 → audio_extract → vad_segment → whisper ffmpeg speech only text out

Put the fleet to work

Tell us what you're building. We'll set up your account, configure your first project with you, and validate it on the fleet before you run at scale.