OpenAI launches Codex-Spark on Cerebras chips for >1000 tokens/sec coding, first beyond-Nvidia inference. Optimized for low-latency real-time collaboration with 128k context. Tradeoffs in complex tasks for speed.
Key Points
- 1.Runs on Cerebras wafer-scale processors
- 2.Near-instant response for coding iteration
- 3.Available in Codex app and VS Code
Impact Analysis
Diversifies OpenAI infra amid Nvidia tensions; boosts developer productivity via responsive AI coding.
Technical Details
Text-only, research preview for Pro users; underperforms full Codex on SWE-Bench but prioritizes latency.
