The latest features, improvements, and fixes across LLM Gateway, DevPass, Lounge, and Airside.
Retain All Data is now an Enterprise feature. New organizations are Metadata Only, and existing organizations that retain full payloads keep the setting until November 8, 2026. Usage analytics and the Responses API are unchanged.
Read more about Full Data Retention Moves to Enterprise →
Call Perplexity's standalone Search API through LLM Gateway at /v1/search. It returns ranked web results with extracted page snippets instead of a model-written answer, billed per request, with failed requests free.
Read more about Perplexity Search API →
Request Azure's low-latency Priority tier with the same service_tier field you already use for OpenAI. The gateway reads back the tier Azure actually served, so a request Microsoft downgrades is billed at standard rates rather than the premium you asked for.
Read more about Azure Priority Processing →
A new `smart` model string: choose which models it may resolve to, and let a classifier rate each request so easy prompts go to a cheap model and hard ones go to a frontier model. Existing `auto` behaviour is unchanged. Available to every organization while in beta; not available on DevPass yet.
Read more about Smart Routing →
End-users who buy credits in your app now get a receipt led by your brand, with your support address and a card statement that reads LLMGTWY* YOURBRAND instead of ours. Refunds send a matching credit note. Available on Payments SDK projects, which are in preview.
Read more about Branded Payments SDK Receipts →
Watch models your compliance policy blocks today and get told the moment a provider that meets your requirements starts serving them — in-app, by email, or in Slack. Downgrade alerts fire when a provider stops meeting the policy. Available on the Enterprise plan.
Read more about Compliance Alerts →
OpenAI's GPT-6 Sol and GPT-6 Luna are live on the gateway at $2 / $10 and $0.10 / $0.50 per 1M tokens — Sol at half the GPT-5.6 Sol rate, Luna at half the GPT-5.6 Luna input rate and 58% below it on output — with cache reads at 10% of input, a 1.05M-token context window, and reasoning from none to max.
Read more about GPT-6 Sol & Luna at Half the Price →
Anthropic's Claude Opus 5.5 is live on the gateway with a 1M context window, always-on adaptive thinking, and $4 / $20 per 1M tokens — 20% below Claude Opus 5 — with cache reads at $0.20.
Read more about Claude Opus 5.5 →
The new /v1/systemone endpoint returns typed decisions with calibrated probabilities instead of generated text. TypeSafe's Jev 1.13 is the first decision model, billed on input tokens only.
Read more about System One: Typed Decisions →
Perplexity retires its Sonar chat completions API on September 27. On September 25 we move perplexity/sonar to Perplexity's Agent API with no changes to your requests and lower prices, and perplexity/sonar-pro and perplexity/sonar-reasoning-pro stop being routable on September 27.
Read more about Perplexity Sonar Changes September 25 →
Showing 10 of 110 updates
Load more