Cerebras's Codex Spark 5.3 has been a huge flop. Small context window and old model. But hopefully they can improve so that we can benefit from 1000 tokens/second with GPT 5.5.