DeepSeek V4 Pro now runs updated weights on AI Gateway
AI Gateway moved DeepSeek V4 Pro’s existing alias to updated weights without a code change. Pinning is documented inconsistently, so verify the dated model ID first.
Calls to **deepseek/deepseek-v4-pro** now receive updated weights by default on AI Gateway, with no application change. The post also offers a dated ID for explicitly selecting the revision in AI SDK or coding-agent configuration.
Existing users should rerun representative agent tasks because an unchanged alias can now produce different behavior. Pin a dated revision when reproducibility matters, and use the gateway’s coding-agent setup when routing a local agent through it.
Calls to **deepseek/deepseek-v4-pro** now receive updated weights by default on AI Gateway, with no application change. The post also offers a dated ID for explicitly selecting the revision in AI SDK or coding-agent configuration. Existing users should rerun representative agent tasks because an unchanged alias can now produce different behavior. Pin a dated revision when reproducibility matters, and use the gateway’s coding-agent setup when routing a local agent through it. The material conflicts on the pinned slug: its prose names **deepseek-v4-pro-0813**, while the code sample uses **deepseek-v4-pro-0831**. Confirm the valid identifier before changing production or evaluation configurations.
This confirms that a stable gateway alias does not guarantee stable model behavior: representative agent evaluations should rerun when weights change, while dated IDs become necessary for reproducibility. The contradictory pinned slugs add a configuration-integrity risk, so neither production pinning nor benchmark attribution is safe until the valid revision ID is verified.