[Private Dinner] Beyond Benchmarks w/ OpenRouter [AI Tinkerers - San Francisco]

[Private Dinner] Beyond Benchmarks w/ OpenRouter

May
14
Thursday
Thursday, May 14th, 2026 6PM to 9PM (PDT)
Address Info
Available on RSVP acceptance

Event Ended

This event has already taken place.

Attendees 20+ registered
Attendees include founders and engineers from Google DeepMind, NVIDIA, and Salesforce, specializing in AI/ML, Python, and data science, alongside an IEEE Rising Star.

(Banner) A promotional graphic for an 'AI Tinkerers' private dinner in San Francisco, featuring a formal table setting against a backdrop of the city's nighttime skyline. Text: AI TINKERERS - SAN FRANCISCO Beyond Benchmarks Evaluating Models in Production ~ Private Dinner ~ MENU openrouter Openrouter Modern, professional, and high-end. | Colors: #050A14, #FFFFFF, #D4AF37 Note: The image is a promotional graphic designed with text overlays to advertise a specific event, functioning as a digital banner.

[Private] Beyond Benchmarks

A Closed-Door Dinner on Evaluating Models in Production

This is not a public event.

On May 14th, AI Tinkerers and OpenRouter are convening a small group of builders for a private VIP dinner in San Francisco. Attendance is by invitation only.

No applications. No +1s. No observers.

Each guest has been hand-selected for their direct experience shipping and operating AI systems in production.


The Theme: When Benchmarks Stop Mattering

Benchmarks are clean. Production is not.

With the rapid iteration of frontier and open models, leaderboard performance has become increasingly disconnected from real-world reliability. The teams actually deploying these systems are forced to build their own evaluation loops-grounded in latency, cost, user behavior, and failure modes that don’t show up in static tests.

This dinner is a closed discussion among those actively navigating that gap.

We will focus on:

  • Live Evaluation Systems: How teams are instrumenting real user traffic to continuously evaluate model performance.
  • Routing as a Primitive: Using gateways to dynamically select, fallback, and optimize across models in production.
  • Model Drift in Practice: Detecting and responding to silent regressions as providers update weights and infra.
  • Cost vs. Reliability Tradeoffs: When cheaper models break-and when expensive ones aren’t worth it.

Format

There are no presentations.

No slides. No intros. No spectators.

Just a single table of builders exchanging what’s actually working-and what isn’t.

  • Hand-Selected Room: Every attendee is actively shipping AI systems at a high level.
  • Off-the-Record: This is a confidential environment. What’s shared in the room stays in the room.
  • Operator-Only Dialogue: We skip strategy and focus entirely on implementation, edge cases, and failure modes.

About the Sponsor

OpenRouter is the largest, most battle-tested AI model exchange: 300+ models from 60+ providers, processing ~30 trillion tokens monthly. Instead of hand-rolling routing logic and juggling provider keys, you route through OpenRouter once and let Auto Router pick the best execution path for every request - with automatic failover when providers degrade.
(Logo) The image features the word 'openrouter' in a bold, lowercase, sans-serif typeface. It is a clean and minimalist wordmark logo. Text: openrouter Colors: #0a0a14 Note: The image is a stylized text-based representation of a brand name, which is a classic form of a wordmark logo.


AI Tinkerers is a curated community of active builders. Events like one this are not open to the public.

Ready for more?

Check out other posts from this blog.

View all posts

Contact Organizers

Questions? We're here to help.