OpenAI's new Spark model codes 15x faster than GPT-5.3-Codex - but there's a catch

OpenAI unveiled GPT‑5.3‑Codex‑Spark, a smaller, latency-first Codex model for real-time, conversational coding that delivers up to 15x faster code generation, 80% lower roundtrip overhead, and 50% faster time-to-first-token, enabled by persistent WebSockets and serving on Cerebras WSE‑3 chips; initially limited to $200/mo Pro users, Spark emphasizes quick, targeted edits and rapid iteration but underperforms the full GPT‑5.3‑Codex on long-horizon agentic benchmarks and does not meet OpenAI’s “high capability” bar for cybersecurity, making it a speed-versus-accuracy tradeoff that users may switch between depending on task and load.