Changelog

What’s new in Gateway-LLM

Curated release notes. For the unfiltered git log, see the GitHub releases.

• v1.4.02026-04-22feature

  Free-first routing mode + simulated multi-provider in the live demo

  • New `free-first` router mode: simple → free Llama, medium → gpt-4o-mini, complex → gpt-4o.
  • Live router on the homepage now routes across 8 models (5 OpenAI + 3 simulated).
  • Routed-model badges in the live router (Free / Cheap / Fast / Smart / Flagship / Reliable).
• v1.3.02026-03-31improvement

  Faster classifier, tighter cache defaults

  • Classifier hot path rewritten — sub-microsecond decisions even on long prompts.
  • Default semantic cache threshold raised from 0.91 to 0.93 (fewer false positives in production).
  • New Prometheus series `gatewayllm_router_failover_total{from,to}`.
• v1.2.02026-02-28feature

  Anthropic + Gemini deployment translation upgrades

  • Anthropic tool-call translation now handles deeply nested schemas correctly.
  • Gemini streaming chunk format converted to OpenAI shape on the server (no client changes).
  • New provider type `openai_compatible` for fine-tuned / self-hosted endpoints.
• v1.1.02026-01-15feature

  Virtual API keys: budgets, scoped routers, signed receipts

  • `monthly_budget_usd` per key with hard 429 ceiling.
  • Per-key router strategy override (`router.force_strategy`).
  • Ed25519-signed audit receipts now ship in the OSS binary, not just the Custom plan.
• v1.0.02025-12-01feature

  1.0 — public release

  • Apache 2.0 licensed.
  • OpenAI-compatible HTTP API across 8 providers.
  • Smart routing with prompt-complexity classifier.
  • Virtual API keys, RPM/TPM limits, exact + semantic cache.
  • Prometheus, OpenTelemetry, Datadog, Langfuse exporters.