Inception Releases Mercury 3 Diffusion LLM at 1,068 Tokens per Second

- Inception dropped Mercury 3, a diffusion LLM hitting 1,068 tokens per second. It’s faster than your typical moonboy’s delusion, claiming to beat GPT-6 Luna and Claude Haiku 5.5 on most benchmarks. The catch? It’s API-only, so you can’t hoard weights like a paranoid data terrorist. While Inception boasts about parallel refinement, remember that speed doesn’t stop a meat wallet from getting rekt by bad code. At $0.20/1M tokens, it’s cheap enough to burn through your stack-eye funds testing agentic tools that probably just hallucinate market crashes. Check the hash manifest for yourself; don’t trust vendor tables any more than you trust Core Dynamics paperwork.