Route by price or speed, one endpoint for realtime, and explore dataset scoring and caching.  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏  ͏ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ ­ 
Respan

Respan Updates

November 18, 2026

Route providers using model suffixes

Add a suffix to any model name and the gateway picks the provider for you.

  • :cheapest orders the providers serving that model by price.
  • :fastest orders them by measured latency.
  • :zdr routes only to providers that support Zero Data Retention (ZDR).

Switch between cost and performance by changing a single suffix. No extra configuration needed.

To get started, open any model on Respan, go to the Providers tab, select Cheapest or Fastest, and copy the generated model ID.

Providers tab on a Respan model page with Cheapest and Fastest routing options

One endpoint for every realtime model

Voice and text sessions run over WebSocket through the gateway. Connect to realtime models from OpenAI, Google and Mistral through one endpoint. The gateway handles protocol translation, so you can switch models without rewriting your client.

Filter the model catalog by Realtime to see everything you can reach.

Respan model catalog filtered to realtime models

Compare prompt versions row by row

Run a prompt or model config against a dataset as a new version, then compare versions row by row to see what actually changed. Score each row with weighted behaviors and get a final score per version, so a regression shows up as a number instead of a wall of output.

Datasets also keep every span after logs expire, support inline editing, and show progress while a run is going.

Two dataset versions compared row by row with per-version scores

See what caching is saving you

Caching now shows up in Metrics: how much time it saved and how many calls hit cache, charted over time. Anthropic prompt caching also hits more often now, because follow-up turns stay on the same deployment.

Metrics chart of cache hits and time saved over time

Also new

  • 🎬 Every modality: Beyond chat, the gateway now handles images, embeddings, speech, transcription and video through the same endpoint.
  • 🧭 Routing in one place: Fallbacks are now set only through routes, so routing config lives in a single spot.
  • 🔑 Workspace provider keys: Share a provider key with selected projects and manage its rotation or removal from the workspace. Custom models work across projects now too.
  • 📇 A cleaner model catalog: Model IDs are now provider/family, with capability icons and the parameters each model supports.
  • 💰 More billing types covered: Video resolution and per-megapixel, page and frame prices are now billed from the catalog.

Read the full changelog →


Respan

​

© 2026 Keywords AI, Inc.
Respan® is a registered trademark.
406 Singleton Ave
Alameda, CA 94501

Don't want to receive this? Unsubscribe