One endpoint. Every model. Zero lock-in.
Your agents just got cheaper.
Point your SDK at oprism.
Set ANTHROPIC_BASE_URL to https://api.oprism.ai. That's it. Every Anthropic-compatible client — the Python SDK, the TypeScript SDK, Claude Code — now routes through oprism. No wrapper libraries, no code changes beyond the env var.
Each request hits the right model.
oprism inspects the request and picks the cheapest provider that meets your latency target. If a provider returns an error, the request retries on the next candidate. Your client sees one stable endpoint; oprism handles failover and provider selection.
You pay less per token.
Because oprism aggregates demand across providers, you get volume pricing without committing to any single one. One bill. One API key. Every model available, no minimum spend.
| Model | Provider | Latency | Cost / 1M tok |
|---|---|---|---|
| claude-opus-4-8 | Anthropic | 842ms | $15.00 |
| claude-sonnet-4-6 | Anthropic | 412ms | $3.00 |
| GCP Vertex | 398ms | $2.74 | |
| claude-haiku-4-5 | Anthropic | 187ms | $0.80 |
| AWS Bedrock | 195ms | $0.72 | |
| deepseek-v4-pro | DeepSeek | 623ms | $1.22 |
| OpenRouter | 701ms | $1.48 |
Every request is logged. You see which provider served each call, latency per route, and cost breakdown by model and provider.
Start building. No credit card. First 1M tokens free every month.
- ›Every Anthropic model
- ›Multi-provider fallback
- ›Request-level telemetry
- ›Usage dashboard + export
ANTHROPIC_BASE_URL=https://api.oprism.ai