Back to changelog

Video Generation, Sessions & More

Generate videos via the API, track conversations with sessions, and more — plus new models and providers.

Video generation and sessions now available on LLM Gateway

Video Generation

LLM Gateway now supports video generation via the API. Generate videos from text prompts using supported models, with job-based async processing and status tracking.

  • Submit a video generation request and receive a job ID
  • Poll for job status — completed jobs return the video URL
  • Failed jobs are properly marked with error details

Read the video generation docs

Sessions

Sessions let you track and organize conversations across multiple requests:

  • Group related requests into named sessions
  • Filter activity logs by session
  • View streaming cost breakdowns per session
  • Improved empty state UI when no sessions exist

New Models

GPT-5.4 Mini & Nano

1openai/gpt-5.4-mini
2openai/gpt-5.4-nano

OpenAI's smaller GPT-5.4 variants are now available — offering the same architecture at lower cost for lighter workloads.

View GPT-5.4 Mini | View GPT-5.4 Nano

MiniMax M2.7

1minimax/minimax-m2.7

New MiniMax M2.7 model mappings added across providers.

View MiniMax M2.7

New Provider: EmberCloud

EmberCloud is now available as a provider, expanding your routing options with additional model coverage.

UI & Platform Improvements

  • Redesigned Code app — Fresh look for the dev plans and coding tools dashboard
  • Revamped admin dashboard — Improved performance on models, mappings, and providers pages
  • Cost simulator revamp — Updated cost simulator with better navigation
  • Activity log filters — Improved filtering in activity logs
  • Status page indicator — Quick link to the status page from the dashboard
  • OpenCode support — Added OpenCode to the integrations list
  • Free icon visibility — Free icon now correctly hidden for paid provider mappings

Documentation & Guides

  • Codex CLI guide — Step-by-step guide for using LLM Gateway with OpenAI Codex CLI
  • Autohand guide — Integration guide for Autohand
  • OpenClaw guide — Integration guide for OpenClaw
  • Introduction page — New docs introduction page
  • Themed screenshots — Light/dark themed screenshots in the learn section

Fixes & Reliability

  • Moonshot thinking disabled for tool calls to prevent errors
  • ByteDance sensitive content blocks now properly classified
  • Cached tokens no longer double-counted in storage cost calculations
  • Bedrock cached tokens correctly included in streaming responses
  • Streaming error diagnostics improved with better logger serialization
  • Canopywave marked as unstable due to availability issues
  • Top-tier model provider compatibility fixes
  • XAI reasoning output option added to configuration