AI vendors decide who gets the model and how
The Gemini 4 Argon thread opened with an EU developer saying the model is unavailable to their org, and linked it to Google's Antigravity tool working badly. The top reaction was "not available. classic google." A related thread covers Anthropic no longer letting Claude Code subscriptions run through OpenClaw. Commenters called it "when you fully use the capacity you paid for, that's too much" and asked for API keys with usage caps instead. One user said they hit rate limits within 1:45 on afternoons.
The pattern here: builders on frontier models are renters. Access depends on region, subscription terms and which harness you use, and all of it can change without warning. That is why the thread asking "Has anyone replaced Claude/GPT with a local model for daily coding?" is getting real answers. One user says Qwen 3.6 27B dense is about Haiku 4.5 level. Another runs DeepSeek V4 Flash on two RTX Pro 6000 cards at 160 tok/s. Others say tokens per second on a Mac M4 is still too low.
The counterpoint is that local setups still lose on speed and tooling. Nobody claims parity with the top cloud models yet. The gap is narrowing for mid-tier work, though.
So what?
If your product depends on one frontier model, your roadmap depends on that vendor's regional rollout and pricing mood. Build a model-agnostic layer now and test an open-weight fallback for the cheaper tasks. EU customers may get new models later than US ones, so plan for that gap.