Sference is live on Opper: European inference for frontier open models

Sference's managed inference is now available through Opper. Six models — Kimi K3, GLM-5.2, DeepSeek V4 Flash, Qwen3.6 35B-A3B, Qwen3-VL 30B, and BottleCap AI's ThinkingCap — are callable through the Opper API today, on a European route with compute inside the EEA and zero data retention on by default.

Opper is the European AI gateway and control plane for agents: one EU-hosted, GDPR-compliant API in front of 300+ models, with routing, fallbacks, observability, guardrails, and OpenAI SDK compatibility. If your stack already goes through Opper, Sference is now one of the providers behind it.

What this changes

Sference becomes a model string, not a procurement cycle. Teams already on Opper reach our models with the same API key they use for everything else. No second account, no second contract, no separate integration. For teams evaluating European inference, that removes most of the reason to postpone the test.

It's a European route, not a European label. Inference and storage of customer content run on infrastructure inside the EEA, on European-owned GPU capacity that sits outside US CLOUD Act jurisdiction architecturally rather than by policy. Zero data retention is the default rather than an upsell, there is no training on customer data, and our DPA is published rather than negotiated one customer at a time. What matters for a compliance review is where the data is processed and retained, so it is worth being precise about it.

Production discipline carries over. Every model is pinned to a specific version, so the checkpoint you evaluated is the checkpoint that serves your traffic, and the serving version is recorded on every request. Realtime, flex, and async completion windows run behind one OpenAI-compatible API on the same GPU fleet, so an interactive agent turn and an overnight backfill hit the same pinned models with the same audit trail.

The gateway does the parts we deliberately don't. Routing across models and regions, real-time PII masking and content filtering, budget caps, spend visibility, and audit trails all live in Opper's control plane. Pin Sference for a task, or set us as the European fallback behind something else. We stay in the infrastructure layer and Opper handles the layer above it.

"We're glad to welcome Sference to Opper. They give our developers exactly what European teams keep asking for: frontier open models served in Europe with zero data retention by default."

— Göran Sandahl, Co-founder and CEO, Opper AI

Models live today

Every Sference route on Opper is European, with compute inside the EEA. Prices are USD per 1M tokens and subject to change; the live catalogue is on the Sference provider page.

Kimi K3 carries the full one-million-token context window with tool calling, structured output, and adjustable reasoning effort. ThinkingCap is BottleCap AI's reasoning-tuned checkpoint. The Qwen 3.6 line is there for the workloads where cost per token decides the architecture — classification, extraction, and enrichment runs that never needed a frontier model. All of them run under the same residency and retention terms.

Get started

If you are already on Opper, switch the model string:

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.OPPER_API_KEY,
  baseURL: "https://api.opper.ai/v3/compat",
});

const completion = await client.chat.completions.create({
  model: "sference/deepseek-v4-flash",
  messages: [{ role: "user", content: "Hello" }],
});
console.log(completion.choices[0].message.content);

Opper is drop-in compatible with the OpenAI, Anthropic, and Google SDKs, so one API key and the model string are all you need. Their quick start covers routing, fallbacks, and evaluations.

If you would rather come to us directly — dedicated deployments, fine-tunes and distillations, larger context and volume commitments, or on-premise — that path is unchanged: start at the docs or sign up.

About Opper AI

Opper AI is the European AI gateway and control plane for agents: one EU-hosted, GDPR-compliant API across 300+ models, with smart routing, automatic fallbacks, built-in evaluations and observability, real-time guardrails, and full OpenAI SDK compatibility. Founded in Stockholm, Sweden.