openroutiq
The self-learning AI router
3 followers
The self-learning AI router
3 followers
The self-learning AI router that picks the best model, provider, and reasoning level for every request - beating OpenRouter Auto, LLMRouter, and Semantic Router, and matching RouteLLM accuracy at 65.1% lower measured cost in benchmarks. - antat-ai/openroutiq
Stop hardcoding AI models to tasks - or sending every request through the same one-size-fits-all default.
The right model changes with quality requirements, cost, latency, complexity, and risk. Your routing layer should adapt too.
Today, we’re launching OpenRoutiQ v0.1.0 - an open-source, explainable, self-learning router that selects the right model, provider, deployment, and reasoning level for every request.
And the benchmark results are strong.
Across our declared evaluations:
• OpenRoutiQ reached 60.43% selection accuracy
• OpenRouter Auto reached 56.09%
• Tested LLMRouter variants ranged from 33.91% to 49.57%
• OpenRoutiQ led all five measured Semantic Router configurations
• It matched RouteLLM’s best accuracy at 65.1% lower measured cost
• It exceeded xRouteBench’s best published macro by 2.63 percentage points
The attached plots show the comparisons, scope, and measured cost-quality trade-offs.
But routing accuracy is only the beginning.
OpenRoutiQ can learn from trusted quality, latency, cost, failure, and drift signals - while keeping automatic promotion under operator control. Every decision is explainable, constraints remain enforceable, and explicit model pins always win.
The core is local, dependency-free, requires no API key, and makes no provider calls. Optional integrations add provider execution, an OpenAI-compatible proxy, LangChain and LangGraph workflows, plus privacy-bounded OTLP, LangSmith, and Langtrace observability.
Install it now:
pip install openroutiq
GitHub: https://lnkd.in/duMt7sph
PyPI: https://lnkd.in/dyNPFxYw
We’d love developers and AI teams to test it against their own workloads, challenge the results, and help build the next generation of intelligent model routing.
If your application still sends every request to one model, it’s time to rethink the routing layer.
hashtag#OpenSource hashtag#ArtificialIntelligence hashtag#LLM hashtag#AIInfrastructure hashtag#MachineLearning hashtag#Python hashtag#MLOps hashtag#GenerativeAI hashtag#LLMRouter
hashtag#Note: OpenRoutiQ is an early public release. We’re sharing it now to learn from real-world workloads, so APIs and configuration formats may evolve before 1.0.
Your feedback, benchmark results, and contributions will help shape what comes next.