One API key to 300+ models, hosted in the EU. Drop-in compatible with the OpenAI, Anthropic, and Google SDKs: switching is a base URL change. What's different: ~half our 30+ providers run inference in Europe, one EU sub-processor covers every model, no prompts stored by default. Add the control plane for routing, PII masking, per-team spend caps, and audit trails. Agent-native: paste one line into Claude Code or Cursor and it sets up Opper for you. No markup on tokens, 3% fee on credit top-ups.
The 3% fee on credit top-ups with no markup on tokens is a cleaner model than most aggregators. For a team with high and variable token usage, is there a minimum top-up size, and does the 3% apply to every top-up regardless of size or does it get negotiated away at enterprise volume? Curious where the unit economics start to favor Opper over going direct to providers.
OpenAI-compatible base URL changes are the easy part. Migration risk usually shows up in streaming, tool calls, response fields, and provider-specific parameters. Which differences does Opper normalize, and which ones do you intentionally reject? A public compatibility table by provider would make the real switching cost much easier to judge.
I like that you’re not just aggregating models but also adding things like PII masking, audit trails, and routing into the same gateway. Was the decision to keep prompts unstored by default mainly driven by privacy requirements in Europe, or did customer feedback push you in that direction?
@amjad_shaik our gateway tier is by design very light and zero data retention because it is very important for some that either need privacy or low cost. On our control plane tier we add options to add tracing with configurable retention, guardrails etc. We hope this strikes a good balance. Thanks for engaging!
Massive congratulations on getting this live @gsandahl qq does the unified schema support advanced provider-specific features like structured outputs or tool-calling seamlessly across different models?
@priya_kushwaha1 yes!
We offer compatability endpoints for Anthropic (messages), OpenAI (chat completions, responses, open responses), Google (interactions) and allow for using the right endpoint for the right model to avoid lossy translations. For many use cases all models can be used across all compat endpoints, but to get the most advanced feature access we recommend using the right endpoint.
Some things we are looking very closely at:
- Caching
- Tool calling
- Structured output
- Provider server side tools
@goran_sandahl Very helpful context makes sense to use the right endpoint for the most advanced provider-specific features while still keeping broad compatibility. Best of luck
Howdy Product Hunt! Co founder Göran here.
One thing we care a lot about is to not just track the frontier, but also the base. We run benchmarks on common tasks and track model performance, cost and time to complete. For example, GLM 5.2 does not perform as well as Fable across these standard tasks, but it runs the benchmarks at 1/14:th of the cost.
Models should do 100% on these benchmarks: https://opper.ai/real-world-benchmarks
Opper AI
@goran_sandahl 100%
Congrats on the launch!
Felix nailed the "lead changing every week" problem. I’ve been using Opper for agentic coding workflows, and being able to route different subagents to different models from one account has been a real time-saver.
Worth checking out if you’re building with multi-agent workflows.
Opper AI
@nileback thanks so much, great to hear from a happy customer 🙏
Switched two prototypes over in under five minutes, just swapped the base URL and it worked. Really appreciate the no markup on tokens, makes it way easier to forecast costs.
@nevzatyolag2y3 great! Let us know if you have any feedback or questions!