Research & Insights
Notes from the platform.
We ship findings, not hype. Latency reports, evaluation notes, and patterns from production — written like OpenAI research, not marketing.
Product · Aug 25, 2026
Intel-4 is now available — frontier reasoning, production-ready
128k context, streaming and function calling. Same API shape, new reasoning benchmark highs. No vendor lock.
Engineering · Aug 18, 2026
How we keep P50 under 50ms at scale
Edge routing, what we cache and what we never cache for privacy. With flame graphs you can verify.
Research · Aug 12, 2026
The model-agnostic playbook
Why serious teams treat models as interchangeable and invest in evaluation, not allegiance.
Security · Aug 05, 2026
Secure by design: hashed keys, no training on your data
How we built abuse detection and rate limiting without ever logging your prompts.
Have a topic you want us to benchmark?
We publish what we measure. Suggest an eval.