OpenAI has launched the GPT-5.6 model family, designed to deliver frontier-level agent performance at significantly lower costs by optimizing reasoning effort and introducing new API controls. The update allows startups to build faster, more capable agents using smaller models like Luna and Terra for high-volume tasks while reserving larger models for complex judgment.
- GPT-5.6 Sol at "low" reasoning outperforms GPT-5.5 at "high" reasoning on Agents’ Last Exam with constant harness conditions.
- On the BrowseComp benchmark, GPT-5.6 Luna scored 84.04% for $1.33, compared to GPT-5.5's 84.36% for $33.27.
- New Responses API primitives include persisted reasoning, native compaction, multi-agent orchestration, and programmatic tool calling.
- Enabling retained reasoning and compaction on ARC-AGI-3 increased GPT-5.6 Sol's score from 13.3% to 38.3% while using roughly 6x fewer output tokens.
- Prompt cache TTL is extended to a minimum of 30 minutes with deterministic cache breakpoints to improve hit rates and reduce latency.
These changes shift the economics of building agents, enabling comparable or better results at a fraction of the cost by using smaller models and efficient architectural choices.