Five new providers: Fireworks AI, Groq, Together AI, xAI, and Z.ai
The catalog doubles to ten providers and passes 200 models — all behind the same key, the same budgets, and the same invoice. Switching is still one string.
The provider catalog just doubled. Alongside OpenAI, Anthropic, Google Gemini, Mistral and DeepSeek, Open Vertex Router now routes to Fireworks AI, Groq, Together AI, xAI, and Z.ai — ten providers, more than 200 models, one API.
Nothing about how you use the platform changes. That’s the point.
What each one brings
- Groq serves open models on its custom LPU hardware, built around very low time-to-first-token — the option to reach for when latency is the constraint.
- Fireworks AI focuses on fast serving of open and fine-tuned models, with a catalog that tracks new open releases closely.
- Together AI carries one of the broadest open-source line-ups available — if a notable open model shipped recently, it’s usually here.
- xAI brings the Grok family.
- Z.ai brings the GLM family, including its coding-focused variants.
These are the providers’ own positions, not our benchmarks. Which one wins for your workload depends on your prompts, your latency budget, and your traffic shape — the platform’s job is making that comparison a config change instead of an integration project.
Switching is still one string
Same endpoint, same key, same OpenAI-compatible request. Moving a workload from one provider to another is the change you’d expect:
{
- "model": "mistral/mistral-large-latest",
+ "model": "groq/llama-3.3-70b-versatile",
"messages": [...]
}No new SDK, no new account, no new invoice line to reconcile.
The controls don’t care which provider you picked
Every mechanism that made the platform worth adopting applies to the new providers on day one, because they sit in the request path, not in a per-provider integration:
- Prepaid budgets and hard caps — a request to Groq is pre-authorized exactly like a request to OpenAI. A runaway loop can’t blow past your limit on a new provider any more than on an old one.
- One invoice — five more providers is zero more invoices.
- Model allowlists — new providers are opt-in per account. Nothing new is reachable until you allow it, so a wider catalog never widens your compliance surface by surprise.
- Usage breakdown — spend and tokens per key, model and provider, so the first thing you learn about a new provider is what it actually costs you.
Trying one
If your account’s allowlist already covers the models you want to try, it’s the one-line change above. If not — or if you’re not on the platform yet — talk to sales: scoping which providers and models fit your workload is exactly what the onboarding call is for.