OpenAI rolled out strict JSON Schema Structured Outputs and function calling for the o3-mini API. Supported by new reasoning kernel optimizations, generation speed reaches 120 tokens/s, allowing coding agents to produce type-safe interfaces and AST payloads with zero JSON syntax drift at low latency.
Key Takeaways
- βGuarantees 100% JSON Schema adherence, eliminating syntax parsing failures during tool use.
- βAchieves up to 120 tokens/s decoding throughput, an 80% speedup over previous reasoning models.
- βOffers low, medium, and high reasoning effort toggles for varied latency and depth needs.