OpenAI's GPT-6.1 Terra model

OpenAI launched GPT-6.1 Terra on October 26, 2026, priced at $1/million input tokens and $5/million output tokens, roughly half the cost of GPT-5.6 Terra and GPT-6.1 Sol, and about 10x the price of the cheaper GPT-6 Luna. The model targets computer use, visual reasoning, coding, and business automation rather than being OpenAI's most powerful offering. On an internal coding benchmark, Terra scored 50.3% versus 21.5% for its predecessor GPT-5.6 Terra. Hallucination error rates fell from 53.6% to 16.7% at low reasoning effort and to 4.3% at maximum reasoning effort. Terra performs on par with flagship GPT-6 Astra on certain professional agent benchmarks while costing one tenth as much per token, and is more than 3x faster than both Astra and GPT-6.1 Sol on those tasks. Separately, a paper published October 22 evaluated GPT-6-Astra across 12 robot benchmarks, finding it nearly saturated the LIBERO suite and scored 81.9% on RoboDojo and 87.3% on DexJoCo, but could not verify physical success, recover from irreversible mistakes, or run affordably in real-world settings. A single LIBERO-Ultra-Long episode consumed approximately 9 million tokens and took nearly 2 hours to complete. The authors concluded general models show research promise as robot controllers but that task-specific data, heavy compute, and physical constraints block near-term deployment.