--- format: "aidr-story-markdown/v1" id: "8cd162122744a01de963ddb5771c30be55736410d40db135a2eed3f2b898c24d" canonical_url: "https://aidr.today/8cd16212?lang=en" title: "OpenAI Targets AGI by End of 2026 With 10 Trillion Parameter Bel Model" lang: "en" requested_lang: "en" available_langs: ["en","vi"] translation_fallback: null fallback_fields: [] published_at: "2026-08-26T14:57:59.000Z" category: "Models" topics: ["openai","sam-altman","agi","llm"] source_urls: ["https://huggingnews.com/ai/openai-targets-agi-by-end-of-2026-with-10-trillion-parameter-bel-model-f0a08382","https://x.com/zephyr_z9/status/2092615182396936586"] summary: "An automated AI research intern codenamed Astra can implement experimental ideas inside a codebase, run experiments, and perform work that previously occupied a human for a week. CEO Sam Altman believes OpenAI will possess an internal system qualifying as artificial general intelligence by the end of 2026. This goal relies on a new pretrain model called Bel, which exceeds 10 trillion parameters and serves as the base for both Astra and GPT-6. The company recently paused its largest frontier reinforcement learning run after evidence suggested Astra may cross a critical cybersecurity threshold. Competitor Anthropic lacks the compute required to launch a competing response to Astra this year, though it expects to regain its lead in early 2027. These training advancements are supported by Project Stargate, which provides the compute capabilities to build more advanced models over the long term." --- # OpenAI Targets AGI by End of 2026 With 10 Trillion Parameter Bel Model > [Open the canonical story]() **Published:** 2026-08-26T14:57:59.000Z **Category:** Models **Topics:** openai, sam\-altman, agi, llm ## Summary An automated AI research intern codenamed Astra can implement experimental ideas inside a codebase, run experiments, and perform work that previously occupied a human for a week\. CEO Sam Altman believes OpenAI will possess an internal system qualifying as artificial general intelligence by the end of 2026\. This goal relies on a new pretrain model called Bel, which exceeds 10 trillion parameters and serves as the base for both Astra and GPT\-6\. The company recently paused its largest frontier reinforcement learning run after evidence suggested Astra may cross a critical cybersecurity threshold\. Competitor Anthropic lacks the compute required to launch a competing response to Astra this year, though it expects to regain its lead in early 2027\. These training advancements are supported by Project Stargate, which provides the compute capabilities to build more advanced models over the long term\. ## Sources - [Story source]() - [Story source]()