Chat Plans, Service Tiers, SDK Sandbox & More
Monthly Chat subscriptions from $9, Flex and Priority service tiers, sandbox test keys for the LLM SDK, a no_training model filter, public DevPass profiles, and a stack of product polish.
A roundup of everything else that shipped recently — new ways to pay, new ways to route, and a safer way to build on the LLM SDK.
Chat Subscription Plans
Chat now has monthly plans, so you can use every studio without topping up credits manually:
- Starter — $9/mo with 2× credits ($18 of usage)
- Plus — $19/mo with 2.5× credits ($47.50 of usage)
- Pro — $49/mo with 3× credits ($147 of usage)
Plus and Pro unlock the full catalog including frontier models; Starter covers everything except the most expensive flagships. Higher tiers earn a more generous multiplier — same philosophy as DevPass.
Flex & Priority Service Tiers
Supported OpenAI and Google models now accept the OpenAI-compatible service_tier parameter:
1{ "model": "google-vertex/gemini-2.5-pro", "service_tier": "flex" }1{ "model": "google-vertex/gemini-2.5-pro", "service_tier": "flex" }flex— about 50% cheaper, best-effort processing for batch and background workpriority— processed ahead of standard traffic when latency matters- The gateway forwards the tier only where the provider/model supports it, and rejects unsupported combinations with a clear error
Sandbox Test Keys for the LLM SDK
Building on the embeddable wallets SDK? You can now create test secret keys alongside live ones. Top-ups made with a test key run through the Stripe sandbox — build and test your whole top-up flow without real charges. There's also a new SDK page under project Settings to manage secret keys, end-user session settings, markup, and allowed origins.
Filter Models by Training Policy
/v1/models now accepts a no_training filter, returning only models whose providers don't train on your request data — handy for compliance-sensitive routing.
Model Categories & Fair-Use Caps
Every model is now categorized as Premium or Standard, powering dashboard filters and analytics. For DevPass coding plans, premium models get a rolling weekly fair-use cap (10/50/140 credits for Lite/Pro/Max) so flagship capacity stays available for everyone — the API and pay-as-you-go credits are unaffected.
How categories and caps work →
Public DevPass Profiles
Claim a username and share your AI coding activity. Your profile shows your activity heatmap and top coding agents at llmgateway.io/profiles/<username> — private by default, shareable when you flip it on, complete with a dynamic OG image.
Product Polish
- Org switcher in the studios — switch organizations without leaving Image, Video, or Audio Studio, and generations are billed to the right org even if you switch mid-flight
- Plain-English legal pages — terms and privacy now open with a human-readable summary
- Delete account in Code settings — DevPass users can remove their account directly from settings
- Cost-aware sticky sessions — sticky session routing now factors cost into provider selection