The operating layer between your product and every LLM.
BestLLM is an OpenAI-compatible LLM API gateway that sits between your application and every model provider. Keep the developer experience simple — one endpoint, one SDK format, one bill — while your team owns provider approvals, regional policy, spend controls, and auditability.
One compatible API for every model
Point existing OpenAI SDK integrations at one stable surface and change model routes without rewriting every application. Your code uses the same chat completions format, the same streaming protocol, the same response shapes — regardless of whether the upstream model is DeepSeek Chat, Qwen3.8 Max, Kimi K3, GLM-5, or MiniMax M3. Adding a new model to your workflow means changing one field in the request body, not onboarding a new provider contract, a new SDK, or a new billing relationship.
Policy enforcement before dispatch
Region, retention, provider, model, and fallback rules are evaluated before a request leaves your boundary. Every model route requires explicit approval — if a model is not on the approved list for a workspace, the request is rejected before it reaches the upstream provider. This is compliance-aware routing for US and EU deployments: you control which regions serve traffic, which models a team can access, and what happens when a provider is unavailable. No cross-region fallback fires unless your workspace policy explicitly allows it.
Commercial truth: ledger you can reconcile
Reservations, captures, releases, price snapshots, and append-only ledger entries turn usage into explainable charges. When a request arrives, the gateway reserves the estimated cost. As the response streams, the reservation captures actual token usage. When the response completes, the final charge is recorded with a versioned price snapshot from the moment of dispatch. Your invoices reconcile to individual requests — not monthly aggregates with unexplained variance. See our pricing page for plan details and the per-model cost breakdown.
Workspace isolation and RBAC
Every API key belongs to a workspace. Workspaces sit inside organizations. Roles control who can create keys, approve model routes, set spend limits, and view billing data. Audit trails record every configuration change. This means your engineering team can self-serve model access within approved boundaries, while your security and finance teams retain oversight. The gateway enforces boundaries at the edge — not in after-the-fact log reviews.
How requests flow through BestLLM
Your application sends a chat completion request to https://api.bestllm.dev/v1/chat/completions with a model ID and workspace API key. The edge service authenticates the key, checks workspace policy (approved models, regional constraints, spend limits), reserves the estimated cost from the wallet, and dispatches to the approved provider route. The provider response streams back through the edge, tokens are captured in real time, and the final ledger entry is written on completion. If anything fails — invalid key, unapproved model, region violation, insufficient balance — the request is rejected before it reaches the provider.
Supported model routes
Every model below is live and accessible through one OpenAI-compatible endpoint. Click through for full pricing, context window specs, benchmark data, and API integration examples.
Also see our Best Chinese AI Models 2026 comparison for pricing and benchmark details across all providers, or read the API documentation to get started in five minutes.