VercelEngineering PostOfficial Source

DeepSeek V4 Flash is 90% off through Novita on AI Gateway

Vercel Pro users can route DeepSeek V4 Flash to Novita for 90% off through August 11. Provider fallback remains enabled, but fallback requests use standard rates.

Vercel · Aug 4, 2026

Open Source Open Markdown Open JSON

Source Summary

Vercel AI Gateway offers **90% off** DeepSeek V4 Flash through Novita for Pro customers until **August 11**. Both the base model and 0731 variant qualify.

Practical Implication

Put Novita first in the provider order and monitor which provider serves each request. Budget agent workloads against the fallback price, not only the promotional rate.

Agent-Ready Context

Vercel AI Gateway offers **90% off** DeepSeek V4 Flash through Novita for Pro customers until **August 11**. Both the base model and 0731 variant qualify.

Put Novita first in the provider order and monitor which provider serves each request. Budget agent workloads against the fallback price, not only the promotional rate.

If Novita cannot serve a request, Gateway falls back to other providers at standard rates. After the deadline, the models remain available at standard rates with no markup.

Connected Context · Feed7 Judgment

This temporarily changes routing economics rather than model capability: Novita-first routing can sharply lower cost only while it serves the request and the promotion remains active. Prior gateway evidence makes provider telemetry, fallback behavior, and task-level evaluation essential; budgets and routing decisions should therefore use the standard fallback rate and be revisited after August 11 rather than treating the discount as a durable model advantage.

AI Gateway: GPT-5.6 pricing and speed updatesBoth change gateway economics without establishing better task outcomes, but the GPT-5.6 reductions persist under unchanged model IDs while the DeepSeek discount is provider-specific and time-limited.diegosouzapw/OmniRouteOmniRoute reinforces the operational consequence of the promotion: quota failover and cost telemetry are needed because the selected low-cost provider may not serve every request and fallback can change the realized price.TRACE-ROUTER: Task-Consistent and Adaptive Online Routing for Agentic AITRACE-Router adds an outcome-based counterweight to promotion-driven routing: a discounted route should still be evaluated at the task level, with backend consistency considered when fallback changes the serving provider.AI Gateway adds unified fast mode supportBoth require inspecting routing metadata rather than assuming the requested option was delivered: fast mode may fall back to standard speed, while Novita failures may fall back to standard-priced providers.

Context Map

infracoding#gateways#model-selection

Uncertainty

If Novita cannot serve a request, Gateway falls back to other providers at standard rates. After the deadline, the models remain available at standard rates with no markup.