New prototype geoclass pools OpenAI subscriptions to bypass regional inference capacity limits
TheZachMueller · x · 2026-10-17
Geoffrey Huntley released geoclass, an experimental prototype that pairs with his month-old underclass proxy to work around OpenAI's region-based inference capacity limits.
- Motivation: With the ongoing compute crunch, the author kept hitting "model is at capacity" errors during European peak hours and suspected capacity is allocated per region — consistent with Anthropic models feeling faster in Australia during US nighttime last year.
- Underclass pools multiple ChatGPT subscriptions behind one proxy, auto-claims resets, and pins the prompt cache key to avoid wasting usage; the architecture includes sticky sessions, a health pool, fail-fast saturation, and prompt cache pinning.
- GeoClass routes requests across inference regions (EU/UAE/US) by discovering Tailscale-tagged VMs deployed worldwide, with a preflight layer that scans requests and attachments, scrubs secrets, and redacts or blocks content.
An engineer-built workaround for regional inference throttling — "works in theory," per the author.
Related event: Developer Pools ChatGPT Subscriptions to Bypass Regional Rate Limits(2 posts)→
More from coding & agent
- One user made 24,000 Copilot requests in 24h: how firms audit 580 AI seats nobody tracks — alex_verem · 2026-10-17
- Sports odds MCP server compares Pinnacle fair odds with Polymarket and Kalshi prices — Unhappy_Composer_894 · 2026-10-17
- Dev adds MCP server to a dance directory, shares Claude vs ChatGPT vs Grok testing notes — Ajunjahi · 2026-10-17
- Palo Alto Networks ships Cortex XCOR: AI SRE agent diagnoses alerts in under 3 minutes — shashib · 2026-10-17
- HF dev begs OpenAI/Anthropic to fix agents picking the wrong Chrome window — NielsRogge · 2026-10-17
- An agent negotiated a 3PL contract in 3 days, end to end — aarthir · 2026-10-17