OpenAI

OpenAI is the US AI lab that makes the GPT family of language models. It announced itself on December 11, 2015 as “a non-profit artificial intelligence research company”.¹ In 2019 it created a for-profit subsidiary, and in a restructuring announced on October 28, 2025 the nonprofit became the OpenAI Foundation and the for-profit became a public benefit corporation, OpenAI Group PBC, which the Foundation controls.² Its API catalog is led by three GPT-6 models: GPT-6 Astra for complex reasoning and coding, GPT-6.1 Sol to balance intelligence and cost, and GPT-6 Luna for cost-sensitive, high-volume work. It also publishes two open-weight models, gpt-oss-120b and gpt-oss-20b, the larger under the Apache 2.0 license.³⁴

On ARC-AGI-3’s Semi-Private set, run by ARC Prize on its Standard harness, OpenAI models hold the top two scores: GPT-6 Astra at 62.71% and GPT-6.1 Sol at 52.73% (see Highest ARC-AGI-3 Scores, by Model).⁵ OpenAI published the SWE-bench Verified coding benchmark in August 2024, stopped reporting it in February 2026 after its own audit found flawed tests and training-data contamination, and in July 2026 estimated that about 30% of SWE-bench Pro’s tasks are broken (SWE-bench Verified, Pro and What Replaced Them).⁶⁷ Its API prices span both ends of Cheapest LLM APIs per 1M Output Tokens: GPT-6 Luna lists $0.50 per 1M output tokens and GPT-6 Astra $50.⁸ On OpenRouter, GPT-6 Luna processed 7.11 trillion tokens in the week to October 12, 2026, seventh among all models (Most-Used AI Models on OpenRouter by Weekly Tokens).⁹

OpenAI’s compute runs through partners. Epoch AI counts seven US Stargate sites with over 9 GW of planned capacity; the first, in Abilene, Texas, uses chips owned by Oracle, and Epoch estimates 421 MW of IT power in operation there (Stargate’s Sites: Operating vs Under Construction vs Planned; Largest AI Data Centers by Estimated Power Draw).¹⁰¹¹ OpenAI Group PBC is one of eight US companies named in a July 2026 export rule that eases licensing for AI chips sent to the UAE (How US AI Chip Export Controls Work).¹² It shut down its Sora 2 video models and Videos API on September 24, 2026 (How AI Video Generation Works),¹³ and AMD says OpenAI expects to bring its Helios racks online beginning in the fourth quarter of 2026 (AI Accelerators With the Most HBM per Chip).¹⁴