GPT-6 Switching to a Larger Base Model, DeepSeek V4 GA May Beat GLM-5.2

AI leaker Leo claims that OpenAI’s GPT-5.6 will be the last model in the 5.x series, with GPT-6 expected to launch in about a month — or even sooner.

GPT-6 will be based on a new, larger pre-trained foundation model, rather than the Spud architecture rumored for GPT-5.5 and GPT-5.6. Spud is said to be around 4 trillion parameters. OpenAI originally planned to use Spud for GPT-6, but reportedly opted for a bigger base instead.

OpenAI is reportedly excited about the new base, believing it’s a better match to compete against Anthropic’s Fable 5 and the upcoming Fable 5.1.

The same leak also mentions that DeepSeek is preparing to release V4 GA, which could match or exceed GLM-5.2. DeepSeek has also started developing an even larger model to take on MiniMax’s planned 2.7-trillion-parameter MiniMax Pro / M3 Pro.