Blog
News, engineering deep-dives, tutorials, and comparison guides on LLM gateways, routing, agents, and EU-hosted AI from the team building Opper.
Highlights
All posts
People against Jev: what 7,500 games of a picture quiz show
Jev Cards is a picture quiz where people race Jev 1.14, a vision decision model, on the same cards. It reached #1 on Show HN and logged 160,000 answers in its first day. Jev won 77% of games, is close to perfect at recognising things, and loses to people on music, chess and logic.
Jose Sabater · Oct 22, 2026OpenAI-compatible API endpoint: one base URL for 700+ models (2026 guide)
How to point the OpenAI SDK, LangChain or the Vercel AI SDK at one OpenAI-compatible endpoint: the base URL, the API key, model ids, and what works through it.
Felix Wunderlich · Oct 21, 2026GPT-6 Luna vs. GPT-6 Luna Decisions: four times faster
We put GPT-6 Luna in a real-time fighting game twice: through OpenAI's Decisions API and as a chat model, against the same opponent on the same seeds. The Decisions version answered four times faster and won 10 of 10 matches. The chat version won 1.
Jose Sabater · Oct 15, 2026Build your own prompt injection classifier with decision models, or turn on a check
How to build a prompt injection classifier in a few lines with a decision model, how Clef, Jev, Microsoft Decision 1 and an LLM compare on 6,500 prompts, and the one-click check that does it for you.
Jose Sabater · Oct 14, 2026EU data residency for AI: how to enforce it on every LLM call
EU data residency for an AI call is three questions: where the model runs, where the provider keeps content and who owns the provider. How the major AI platforms offer it, where per-call EU switches leak, and how to enforce it on every call with one rule.
Felix Wunderlich · Oct 12, 2026Opper AI partners with Greenference for low-cost EU-hosted inference on small open-weight models
Opper AI partners with Greenference, the French company building off-grid solar datacentres for AI, to add low-cost EU-hosted routes to Qwen3.8 27B, Gemma 4 31B, GLM-4.7-Flash, gpt-oss-20b and Qwen3.6 27B, with prompts and completions never stored.
Felix Wunderlich · Oct 7, 2026GDPR-Compliant AI: How to Use LLMs Under the GDPR (2026 Guide)
What GDPR-compliant AI actually requires when you call an LLM API: who is controller, processor and sub-processor, where inference runs, what providers retain, how transfers to the US work in 2026, and why a model swap is a sub-processor change. With a sourced comparison of OpenAI, Anthropic, Azure, AWS Bedrock, Google Vertex AI and Mistral.
Felix Wunderlich · Oct 6, 2026Run Models at Half the Price with Flex and Priority Tiers
Use OpenAI flex processing, Gemini flex and priority, and xAI priority through one API: set service_tier for half-price or faster tokens, with fallbacks.
Jose Sabater · Oct 5, 2026Opper AI partners with GreenPT for renewable-powered open-weight inference in Europe
Opper AI partners with GreenPT, the Dutch company that builds AI around its energy footprint, to add renewable-powered European routes to Kimi K3, GLM-5.3, GLM-5.3-Flash, DeepSeek V4.1 Flash and MiniMax M2.5, with no training on customer data.
Felix Wunderlich · Oct 1, 2026Opper AI partners with Inceptron for price-performance on EU-hosted inference
Opper AI partners with Inceptron, the Lund company that tunes inference for price-performance, to add EU-hosted routes to the GLM-5.3 pair, GLM-5.2, DeepSeek V4 Flash, and the Kimi K2 line, with API payloads not stored after processing by default.
Felix Wunderlich · Sep 22, 2026EU vs EEA for AI compliance: which countries count and where your model actually runs in 2026
For an AI workload, 'EU data residency' hides five questions: where inference runs, where anything is stored, who operates the endpoint, who made the model, and what is retained. The EU has 27 members, the EEA adds Norway, Iceland and Liechtenstein, and the GDPR follows the EEA line. Which countries count, what each vendor's 'EU region' includes, and how to set the rules per route.
Felix Wunderlich · Sep 21, 2026Demystifying ZDR: what zero data retention actually means
Everyone asks for zero data retention and means something different by it. What the term covers, how a provider actually handles your data, whether you need it, and how we route on terms.
Göran Sandahl · Sep 16, 2026How to run n8n AI agents on EU-hosted models (2026)
Use any model in the n8n AI Agent node with one OpenAI-compatible credential: Claude, GPT, Mistral and open-weight models on EU infrastructure. Install the verified Opper Chat Model node, restrict the dropdown to EU-hosted models, and keep inference and data residency under control. A European alternative to the OpenRouter node.
Göran Sandahl · Sep 7, 2026Opper AI partners with Regolo to bring green inference with zero data retention to the gateway
Opper AI partners with Regolo, Seeweb's inference platform, to bring open-weight models served from Seeweb's European datacenters running on renewable energy to the Opper gateway: GLM-5.2, GPT-OSS, Qwen, Llama, Gemma, and Mistral, plus the fully open Apertus 70B, with zero data retention by default.
Felix Wunderlich · Aug 31, 2026VS Code BYOK: run Copilot Chat on any model (2026)
How to bring your own models to VS Code chat and agent mode in 2026: the built-in Custom Endpoint route, provider extensions, and Opper's 700+ model picker.
Jose Sabater · Aug 27, 2026Opper AI partners with Nextbit to open a Spanish route on the gateway
Opper AI partners with Nextbit, the Valencia-founded AI cloud, to add the gateway's first Spanish route: open-weight models served from Nextbit's own data center in Spain, including ALIA 40B, the language model built by the Barcelona Supercomputing Center for Spain's public administration.
Felix Wunderlich · Aug 20, 2026OpenRouter vs LiteLLM: hosted gateway or open source? (2026)
OpenRouter vs LiteLLM in 2026: hosted marketplace or self-hosted proxy? We compare pricing, routing, observability and data residency, sourced from vendor docs.
Felix Wunderlich · Aug 18, 2026LLM rankings in 2026: how leaderboards work and which to trust
How LLM rankings are built in 2026: Artificial Analysis indices, Arena's blind votes, static benchmarks and vendor scores, what each misses, and which to trust.
Felix Wunderlich · Aug 17, 2026What is an LLM gateway? The 2026 guide
What an LLM gateway is, how it differs from a router or proxy, when you need one, and how to choose between build, buy and open source, updated for 2026.
Felix Wunderlich · Aug 16, 2026Vercel AI Gateway vs OpenRouter: which to pick in 2026
Vercel AI Gateway vs OpenRouter compared on pricing, model catalogs, prompt retention and EU residency, every claim checked against vendor docs in August 2026.
Felix Wunderlich · Aug 15, 2026OpenRouter down? How to fail over in 2 minutes (2026)
The calm OpenRouter outage playbook: read the status signals, tell a 502 from a 503, and fail over to a second OpenAI-compatible gateway in about two minutes.
Felix Wunderlich · Aug 14, 2026Codex router: run OpenAI Codex CLI on any model (2026 guide)
How to route OpenAI Codex CLI to other models in 2026: the config.toml mechanism, why old wire_api tutorials broke, plus tested gateway and local setups.
Felix Wunderlich · Aug 13, 2026Claude Code Router: three ways to run Claude Code on any model (2026)
The complete guide to routing Claude Code beyond Anthropic: the raw env-var mechanism, the open-source claude-code-router, DIY proxies, and the hosted one-command alternative.
Felix Wunderlich · Aug 11, 2026How to add secure, GDPR-compliant AI to your Lovable app
How to add AI to your Lovable app the governed way: route it through the Opper gateway with one server-side secret, EU-hosted models, spend caps, and guardrails, set up by pasting one prompt.
Göran Sandahl · Aug 11, 2026Opper AI partners with Sference to add European inference for frontier open models
Opper AI partners with Sference to bring European inference for open-weight models to the Opper gateway: Kimi K3, GLM-5.2, DeepSeek V4 Flash, Qwen 3.6, and ThinkingCap on EEA infrastructure, with zero data retention by default and pinned model versions.
Felix Wunderlich · Aug 11, 2026Reference images, first and last frames in new video models
MiniMax H3 (Hailuo 3), Grok Imagine Video 1.5, Gemini Omni Flash, Wan 2.7 and Kling 3.0, given the same reference image and the same prompt through one multimodal API. Two characters with a shot list for an anime short, then start and end frame control. What a reference image actually carries, which model follows direction, how long each takes and what it all cost.
Jose Sabater · Aug 6, 2026The 5 best European AI gateways in 2026
Five European AI gateways compared with sourced claims: headquarters, hosting, EU inference residency, retention and pricing for Opper, Cortecs, Requesty, Orq.ai and Eden AI.
Felix Wunderlich · Aug 5, 2026The 10 best AI gateways in 2026
The 10 best AI gateways of 2026 compared on pricing, EU data residency and governance, from Kong and LiteLLM to Envoy, Azure API Management and Opper, plus how to choose.
Felix Wunderlich · Aug 4, 2026EU AI Act Article 50: what product builders have to do now
Article 50 of the EU AI Act applies from 2 August 2026, and it lands on the companies building on models rather than the labs training them. A guide organised by what you ship: chatbots, generated images, generated text, editing features, published AI text. What each requires, which exemptions apply, and how to demonstrate compliance.
Göran Sandahl · Aug 4, 2026The 10 best LLM gateways in 2026
The 10 best LLM gateways of 2026 compared on pricing, EU data residency and production readiness, from OpenRouter and LiteLLM to Vercel, Portkey and Opper.
Felix Wunderlich · Aug 3, 2026Opper AI partners with Novita to bring 90+ open-weight models to the gateway
Opper AI partners with Novita to bring one of the broadest open-weight catalogues to the Opper gateway: 90+ models from DeepSeek, Moonshot, Z.ai, MiniMax, Meta, Qwen, Google, and more, on an OpenAI-compatible API at competitive prices.
Felix Wunderlich · Jul 13, 2026Opper AI partners with Morph to add speed-optimised inference for coding agents
Opper AI partners with Morph to bring its speed-optimised inference for coding agents to the Opper gateway: fast open-weight models including DeepSeek V4 Flash, GLM 5.2, Qwen, and MiniMax, on an OpenAI-compatible API with no training on customer data.
Felix Wunderlich · Jun 26, 2026High-availability inference: same model, every provider
Call a model by its plain name like claude-sonnet-4-5 and Opper serves it across every provider and region that runs it, with automatic failover, region policy, and session-sticky routing. No provider lock-in.
Göran Sandahl · Jun 23, 2026Opper AI partners with Geodd to add European inference, hosted in Norway
Opper AI partners with Geodd to bring European open-weight inference to the Opper gateway: a Norway-hosted route alongside a US route, on custom-tuned GPU infrastructure with zero data retention.
Felix Wunderlich · Jun 17, 2026Universal web search: give any model the web, on any endpoint
Give any model, Mistral, DeepSeek, Llama and more, live web search through one OpenAI-compatible tool. EU-hosted, GDPR-friendly, no search loop to build.
Jose Sabater · Jun 3, 2026We gave Hugging Face's Reachy Mini robot a brain in one hour
How the Opper team turned a Hugging Face Reachy Mini into a talking, moving robot in an hour at MagasinX, Stockholm. A realtime voice AI model through the Opper gateway gave it sight, speech and motion. A hands-on take on physical AI and embodied agents, and why realtime is the easiest way to add an AI brain to any robot or interface.
Jose Sabater · May 29, 2026Opper AI partners with Infercom for sovereign LLM inference
Opper AI partners with Infercom to add sovereign LLM inference to the Opper AI gateway. Open-weight models on SambaNova RDU dataflow hardware, available across EU, US, and JP regions.
Felix Wunderlich · May 12, 2026How to run Claude Cowork with third-party inference providers
How to connect Claude Cowork to a third-party inference gateway like Opper, OpenRouter, or LiteLLM. Setup, model picking, EU residency.
Göran Sandahl · May 4, 2026LLM Router Latency Benchmark 2026: OpenAI Direct vs Router APIs
Do LLM routers add latency? Does region beat model tier? We benchmarked OpenAI direct, OpenRouter, Opper, and 10+ inference backends in 2026.
Göran Sandahl · Apr 21, 2026Introducing Opper Skills for AI Coding Agents
Opper Skills bring the full power of the Opper platform directly into your AI coding agent. Install once, and your agent knows how to use Opper.
Jose Sabater · Jan 30, 2026How we reduced agent costs by 98.6% using Opper context features
A technical deep-dive into how Opper AI context features enabled a small open-weight model to match the quality of a large frontier model, at a fraction of the cost and with faster response times.
Göran Sandahl · Jan 22, 2026Automating In-Context Learning using Feedback
Turn your agent behavior into high-quality training data automatically. Learn how Opper's new feedback system creates a self-improving loop for your LLM applications.
Jose Sabater · Dec 4, 2025Introducing Model Interaction Protocol (MIP) - a Spec for Decoupling AI What from AI How
Introducing Model Interaction Protocol (MIP) - a task specification that decouples what we want models to do from how they should do it, enabling scalable, reliable AI agents in production without constant maintenance.
Göran Sandahl · Nov 25, 2025Introducing Opper Agent SDKs - Framework for reliable, headless agents
Introducing Opper Agent SDKs - powerful Python and TypeScript frameworks for building intelligent, reliable, headless agents with seamless MCP integration, multi model support, comprehensive hooks, and built-in observability.
Göran Sandahl · Oct 14, 2025Keep Your LLM Apps Running: How Opper Handles Fallbacks and Aliases
Opper makes LLM calls more reliable with built-in fallbacks, aliases, and observability—so your apps stay stable without extra code.
Mattias Lundell · Sep 30, 2025Opper AI Acquires FinetuneDB to Accelerate Reliable AI Agents at Scale
Opper AI has acquired FinetuneDB and will integrate its fine-tuning workflow into Opper’s reliability stack to accelerate reliable, task-specific agents.
Göran Sandahl · Sep 11, 2025Opper AI partners with Berget AI to bring EU-hosted serverless inference to one API
Opper AI partners with Berget AI to bring EU-hosted serverless inference to one API
Opper AI · Sep 3, 2025Lost in Context: Where LLMs Shine — and Fail — at Using What They’re Given
We present results of our Benchmark for Context tasks, with insights into what is easy and hard for each of the models
Jose Sabater · Aug 1, 2025Introducing Opper Taskbench - A Real‑World Benchmark for Task‑Oriented LLMs
We built TaskBench to measure real-world LLM performance on practical tasks like RAG, SQL generation, and agentic workflows. Here are our findings across accuracy, cost, and model size.
Jose Sabater · Jul 15, 2025Reference‑Free LLM Evaluation with Opper SDK
Three reference‑free evaluators to demonstrate how to evaluate RAG systems at runtime without gold references.
Mattias Lundell · May 6, 2025Building a Simple GitHub PR Review Agent with ReAct
In this post, we will build an initial version of a simple but effective GitHub PR review agent using the ReAct pattern.
Johnny Chadda · Apr 8, 2025Indexing docs and websites using Github Actions
This guide explores how to automatically index your documentation in Opper Indexes using GitHub Actions for integration with your CI/CD pipelines
Johnny Chadda · Mar 11, 2025Introduction to Schema Based Prompting: Structured inputs for Predictable outputs
Schema based prompting is a technique that allows for instructing models through clear data structures instead of natural language prompts. This post introduces the concept and shows how to use it with Opper.
Göran Sandahl · Mar 4, 2025Introducing Opperator: A composable agent to automate tasks on the web
Opperator is a programmable, autonomous web agent for automating tasks on the web built with the Opper SDK.
Göran Sandahl · Feb 7, 2025Reason then respond with DeepSeek-R1 and Mistral Tiny
Using a reasoning model to generate detailed thought traces that improve the quality of AI responses, while keeping costs low.
Göran Sandahl · Jan 27, 2025Buy to build: an enterprise case for renting the AI gateway
The objective of an enterprise AI program is not to own a gateway — it is to learn, fast, how AI drives business outcomes, while the right architecture is still unsettled. That makes optionality and time-to-value the assets to protect. We make the analyst case for buying the model-access layer and spending your engineering on the differentiated work above it.
Göran Sandahl · Jan 15, 2025Agentic customer service chatbot with tools, tracing and evals
We build a chatbot to assist users by utilising tools, while maintaining context and handling errors gracefully.
Göran Sandahl · Nov 27, 2024Using o1-preview and o1-mini with RAG and structured output
In this blog post we explore how OpenAIs reasoning models o1-mini and o1-preview perform in a RAG pipeline with structured output
Göran Sandahl · Sep 15, 2024Takeaways from AI Engineer World Fair, San Fransisco 2024
Three days at the AI Engineering World Fair in San Francisco, covering the what, how and why of LLMs and how to best use them.
Göran Sandahl · Jun 27, 2024Using Examples and Few Shot Retrieval to Shape LLM Responses
We build a pipeline to shape the output of LLM calls with synthetic examples and few shot retrieval, and see how multiple non-frontier models perform
Göran Sandahl · Jun 20, 2024Resilient Azure OpenAI using Azure API Management
Exploring how to set up APIM with an OpenAI-compatible API, and how to connect it to multiple OpenAI deployments. We will also cover how to set up failover and load-balancing.
Johnny Chadda · Jun 7, 2024RAG metrics: answer correctness
In this blog post we explore how to implement and use RAG metrics to evaluate the quality of answers generated by a model. We will use the answer correctness metric to evaluate the quality of answers generated by a model.
Mattias Lundell · May 28, 2024Introducing Delvin: State of the art bug fixing agent
Delvin is an agent fixing issues from the SWE-Bench Lite dataset, achieving state-of-the-art accuracy(23%) with very simple code leveraging the Opper SDK.
Alexandre Pesant · May 21, 2024Extracting recipes from images using gpt-4o in Opper
In this blog post we explore how to use multimodal models in Opper. We will use the newly released model gpt-4o to generate structured data from images
Mattias Lundell · May 15, 2024Simple RAG with citations
In this blog post we show how to build a simple RAG feature with citations, using Opper with structured input/output and Mistral-Large.
Göran Sandahl · May 9, 2024Examples are all you need: getting the most out of LLMs part 2
Getting GPT 3.5 Turbo to 80% accuracy on the GSM8k benchmark by leveraging the teacher student pattern, synthetic examples and few shot retrieval.
Alexandre Pesant · Apr 30, 2024Getting the best out of LLMs, part 1
Dramatically improving LLMs accuracy with structured generation and chain of thought
Alexandre Pesant · Apr 12, 2024Introducing Opper
At Opper, our mission is to accelerate adoption of Generative AI by making it simpler to build production grade reasoning applications, agents and features.
Göran · Apr 11, 2024