PLATFORMS Signal 626 2 feeds carried it
DeepSeek V4 Pro 0813 weights become default on AI Gateway
↗
DeepSeek V4 Pro's 0813 weight update is now the default routing target on Vercel's AI Gateway, so existing calls pick up the new weights without code changes.
Engineers already calling deepseek/deepseek-v4-pro through AI Gateway get the new weights on their next request with no action required, which is convenient but also means a silent model swap. Vercel positions AI Gateway pricing as pass-through with no markup and no platform fee, including on Bring Your Own Key traffic, so cost-sensitive teams can compare against direct provider billing without a hidden surcharge.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
The unversioned deepseek/deepseek-v4-pro identifier on AI Gateway now routes to the 0813 weight snapshot, picking up automatically for existing code.
To pin to the new version explicitly, set the model to deepseek/deepseek-v4-pro-0813 in the AI SDK or in a coding agent's model configuration.
AI Gateway reflects provider pricing with no markup and charges no platform fee on inference, including on Bring Your Own Key requests.
THE READ
What the cluster adds up to.
What changed. AI Gateway now routes the unversioned deepseek/deepseek-v4-pro identifier at the 0813 weight snapshot, so traffic already pointed at the gateway shifts to the new build without a client-side change. The article does not state which identifier, if any, still maps to the previous weights, or whether the older build remains available at all, so a team that needs to roll back has to find that out from the model leaderboard rather than from this post.
Adoption cost. For teams already calling the unversioned identifier the migration is zero, same model ID, same code, new weights behind it. For teams that want to be deliberate, the AI SDK line is a one-character change: model becomes deepseek/deepseek-v4-pro-0813, and the same string can be set inside a coding agent's model configuration after running vercel ai-gateway coding-agents setup. The post gives no guidance on A/B testing old versus new weights, which is the kind of thing you would want before silently swapping a model in production traffic.
Pricing and BYOK. Vercel repeats the claim that AI Gateway reflects provider pricing with no markup and charges no platform fee on inference, and explicitly extends that to Bring Your Own Key requests. That is the part of the announcement that competes with direct provider billing rather than the weight update itself; an engineer comparing costs needs to know whether their key is BYOK and whether any per-request overhead appears beyond the provider list price. The post does not provide numbers, so any cost comparison still has to be done against the underlying provider.
How the feeds framed it. Vercel's headline is framed as a platform event ("now runs updated weights on AI Gateway"), which is accurate but generic and obscures the fact that there is a specific version behind the change. The Hacker News thread is titled after the specific identifier, "DeepSeek V4 Pro 0813", which signals that the conversation there is about the weights and what is different about them, not about the gateway routing. The HN material on hand is only a comments summary, so the substantive claims to anchor on come from Vercel's post, and any benchmark or regression discussion is not in the provided material.
Written by elseif from the cluster below · checked for specifics the sources never containedTHE CLUSTER