Rumor: DeepSeek Trained a 3T-Parameter Model but Held Back Release

Leaker teortaxesTex claims DeepSeek once trained a 3T-parameter backbone model but shelved its release over cost concerns. He extrapolates that V4.1 Pro could be a 3.1T-parameter MoE activating 30/60B per token with roughly 2.6TB on disk, provided the scaling issues of the V4-Pro stage do not recur.

2026-09-10 ~ 2026-09-10 · 3 related posts

• Speculation: DeepSeek V4.1 Pro could be a 3.1T-param MoE with 2.6TB disk footprint — teortaxesTex · 2026-09-10
• Follow-up: a 3T-parameter model may already exist, scaling issues remain the wildcard — teortaxesTex · 2026-09-10
• Rumor: DeepSeek trained a 3T-parameter model but held back over serving economics — zephyr_z9 · 2026-09-10