DeepSeek V4 Pro and Flash are now available on Vercel AI Gateway

What changed

Vercel added DeepSeek V4 Pro and V4 Flash to the AI Gateway. Both models support a 1 million token context window and are accessible through the gateway's unified API alongside existing providers, with the standard observability, rate limiting, and key management Vercel already exposes.

Why it matters

For developers, this collapses integration cost. You can route a percentage of traffic to DeepSeek V4 without changing application code, just gateway config. That makes real $/M token comparisons cheap to run: shadow the new models, compare outputs and latency on production-like prompts, then shift weight when the numbers justify it. The 1M token window is the headline capability for codebase-scale analysis, long document ingestion, and large RAG bundles where chunking strategies have been hurting answer quality.

What to watch for

Validate tool-calling and structured output reliability before you treat V4 Pro as a drop-in for your current reasoning model; MoE models occasionally regress on strict JSON schemas. Use the gateway's built-in metrics to catch tail latency, especially on prompts above 200K tokens where MoE routing overhead shows up. Set explicit per-model budgets in the gateway and watch your token spend daily for the first week, since aggressive context use can offset the lower per-token price.