#OpenAIReportedlyCompletesBelModelPretraining đš Update: OpenAI already finished the model after Astra
Codename: Bel.
Itâs the successor to Doug.
Doug is expected to be the base for Astra / GPT-6 after more RL.
Bel is a 10T+ pretrain - around the same size class as GPT-4.5.
OpenAI is treating it as a post-GPT-6 base, and some reports even frame it as a possible AGI-threshold foundation.
Thatâs the part people are missing.
Astra is the near-term launch.
Bel is the much larger stack sitting behind it.
OpenAI also thinks Anthropic doesnât have a real public answer to Astra this year.
Mostly compute.
Anthropic is reportedly bracing for the rest of 2026 to go OpenAIâs way, then trying to come back early next year.
Stargate is paying off - not just for inference, but for training the next wave.$KORU $BMT $APR
Codename: Bel.
Itâs the successor to Doug.
Doug is expected to be the base for Astra / GPT-6 after more RL.
Bel is a 10T+ pretrain - around the same size class as GPT-4.5.
OpenAI is treating it as a post-GPT-6 base, and some reports even frame it as a possible AGI-threshold foundation.
Thatâs the part people are missing.
Astra is the near-term launch.
Bel is the much larger stack sitting behind it.
OpenAI also thinks Anthropic doesnât have a real public answer to Astra this year.
Mostly compute.
Anthropic is reportedly bracing for the rest of 2026 to go OpenAIâs way, then trying to come back early next year.
Stargate is paying off - not just for inference, but for training the next wave.$KORU $BMT $APR
