B.AI Upgrades DeepSeek V4 to 0813

B.AI’s zero-migration rollout makes a major model refresh feel operationally painless, but developers should still retest production workloads against the new checkpoints.

• –DeepSeek V4 supports a 1-million-token context window and up to 384K-token outputs.
• –V4-Pro uses a 1.6T-parameter MoE architecture with 49B active parameters, while V4-Flash activates 13B of 284B parameters.
• –Flash offers a lower-cost option for high-volume inference, while Pro targets demanding coding and reasoning workloads.
• –Existing API integrations can continue using the same base URL and model-selection workflow.
• –The quiet release highlights how model upgrades are increasingly becoming infrastructure events rather than headline launches.