DeepSeek V4 Flash now runs updated weights on AI Gateway
DeepSeek V4 Flash jumped 25.8 points on agentic benchmarks. Vercel's AI Gateway now routes to the updated weights by default — no code changes needed.
Why it matters
DeepSeek's stronger agentic model is live in production infrastructure. Practitioners using Vercel's AI Gateway get the upgrade automatically; agents built on the platform pick up better reasoning at no code cost.
The key facts
13 to knowDeepSeek V4 Flash Terminal-Bench score: 82.7 (up from 56.9 in April preview, +25.8 point jump)
Updated weights now live on Vercel AI Gateway by default
Model ID unchanged (deepseek/deepseek-v4-flash) — existing code routes to new weights automatically
DeepSeek only provider serving updated weights initially; others with Zero Data Retention coming next week
Agentic capabilities specifically highlighted as improved
Integration path: use vercel ai-gateway coding-agents setup to connect agents to new model
AI Gateway pricing: reflects provider rates with no markup, no platform inference fee
DeepSeek V4 Flash Terminal-Bench score: 82.7 (up from 56.9 in April preview)
+25.8 point improvement in agentic capabilities
Updated weights now default on Vercel AI Gateway with no code changes required
DeepSeek currently exclusive provider; others (including zero-retention options) coming next week
AI Gateway charges no platform fee, reflects provider pricing with no markup
Integration: agents connect via vercel ai-gateway to auto-route to new weights
Go to the source
Vercel Blogvercel.com
Publisher excerpt: now runs on updated weights by default on AI Gateway, with notably stronger agentic capabilities. On Terminal-Bench, it scores 82.7, up 25.8 points from 56.9 in the April preview.DeepSeek V4 Flash Requests to pick up the new weights automatically, with no change to the model ID or your…
