SaidwireDevelopers
Dashboard Get an API key
OverviewDemosIntegrationsDocsPricingBenchmarksDashboardGet an API key

Changelog

What's new in the API.

Changes to the Saidwire speech-to-text API and its docs, newest first. Changes that break existing clients get at least 14 days' notice by email, as our API terms say.

November 2, 2026

  • New status page with a live check of the API every 5 minutes, uptime and the last 24 hours.
  • New security page: data handling, encryption, keys, where the API runs and our subprocessors.
  • The benchmarks page now shows the full October 29 to 30 run: results by dataset, hour by hour, and the exact method.
  • The API reference now lists every parameter we accept but ignore, and what to use instead.
  • New /llms.txt: a plain-text summary of the API for AI coding tools.
  • This changelog.

November 1, 2026

  • OpenAI-compatible file transcription: POST /v1/audio/transcriptions works with the official OpenAI SDKs. Set the base URL to https://api.saidwire.com/v1 and the model to saidwire-en.
  • The API now runs on two servers in US East with automatic failover between them. It is always on: no wake-up wait.
  • Packages, all version 0.1.0: pipecat-saidwire, livekit-plugins-saidwire and saidwire-mcp on PyPI; @saidwire/ai-sdk-provider and n8n-nodes-saidwire on npm. See Integrations.
  • New saidwire.com, with a quickstart, a pricing page, integration guides, a live speed test and a voice-agent demo.

October 30, 2026

  • Model saidwire-en version 2026-10-30. The version is reported in model_info in every response's metadata.
  • Finished a 24-hour run of Coval's open-source benchmark method: 912 clips, 0 failures. Results are on the benchmarks page.

October 29, 2026

  • Free trial: 120 audio hours, no card and no deadline, with up to 4 streams at once.
  • Self-serve sign-up and API keys in the dashboard, with emails when 80% and 100% of the free hours are used.

October 28, 2026

  • api.saidwire.com launched: real-time streaming over WebSocket at /v1/listen, compatible with Deepgram's streaming API, and pre-recorded files with POST /v1/listen.
  • Finalize is answered right away, without waiting for a pause. The EarlyWord extension sends each word as soon as it's recognized.
  • Telephone audio as it arrives: 8 kHz μ-law and A-law, plus linear PCM from 8 to 48 kHz and common compressed formats.
  • $0.30 per audio hour, billed per second.

Questions about a change? support@saidwire.com.