DeepSeek V4 Flash Gets 90% Off
Cut DeepSeek V4 Flash inference costs by 90% temporarily through Vercel AI Gateway.
30-Second TL;DR
What Changed
Vercel Pro customers receive a 90% discount on DeepSeek V4 Flash through Novita.
Why It Matters
The promotion can materially reduce inference costs for teams already using Vercel AI Gateway and DeepSeek V4 Flash. Developers should account for fallback requests being billed at standard rates and for the discount's limited duration.
What To Do Next
Run a representative workload on Vercel AI Gateway with deepseek/deepseek-v4-flash-0731 and Novita first, then verify fallback costs before August 11.
Key Points
- •Vercel Pro customers receive a 90% discount on DeepSeek V4 Flash through Novita.
- •Use the model ID deepseek/deepseek-v4-flash-0731 and set Novita first in the provider order.
- •If Novita cannot serve a request, AI Gateway falls back to other providers at standard rates.
- •After August 11, DeepSeek V4 Flash remains available at standard rates with no markup.
Deep Insight
AI-generated analysis for this event — not the original article.
Enhanced Key Takeaways
- •Novita AI has positioned itself as a key infrastructure partner for Vercel's AI Gateway, focusing on providing low-latency access to Chinese-developed LLMs for global developers.
- •The DeepSeek V4 Flash model utilizes a Mixture-of-Experts (MoE) architecture optimized for high-throughput, low-cost inference tasks.
- •Vercel's AI Gateway implementation allows for 'provider routing,' enabling developers to switch between model providers programmatically without changing application code.
- •DeepSeek V4 Flash (0731 version) is specifically optimized for long-context window tasks, distinguishing it from previous iterations in the V3 series.
- •The partnership reflects a broader trend of AI model providers using aggressive discounting via third-party aggregators to capture market share from established US-based model providers.
Competitor Analysis
- DeepSeek V4 Flash
- MoE
- GPT-4o-mini
- Dense/Hybrid
- Claude 3.5 Haiku
- Dense
- DeepSeek V4 Flash
- High-throughput/Cost-sensitive
- GPT-4o-mini
- General Purpose
- Claude 3.5 Haiku
- Coding/Reasoning
- DeepSeek V4 Flash
- Aggressive Discounting
- GPT-4o-mini
- Standard Tiered
- Claude 3.5 Haiku
- Standard Tiered
- DeepSeek V4 Flash
- Vercel/Novita/DeepSeek
- GPT-4o-mini
- OpenAI/Azure/Various
- Claude 3.5 Haiku
- Anthropic/AWS/GCP
| Feature | DeepSeek V4 Flash | GPT-4o-mini | Claude 3.5 Haiku |
|---|---|---|---|
| Architecture | MoE | Dense/Hybrid | Dense |
| Primary Use Case | High-throughput/Cost-sensitive | General Purpose | Coding/Reasoning |
| Pricing Model | Aggressive Discounting | Standard Tiered | Standard Tiered |
| Provider Availability | Vercel/Novita/DeepSeek | OpenAI/Azure/Various | Anthropic/AWS/GCP |
Technical Deep Dive
- Model Architecture: Employs an advanced Mixture-of-Experts (MoE) framework to reduce active parameter count per token, significantly lowering inference latency.
- Context Window: Supports extended context lengths, specifically tuned for large-scale document processing and code repository analysis.
- Routing Logic: Vercel AI Gateway uses a weighted round-robin or failover mechanism; the 'Novita first' configuration prioritizes the discounted endpoint before falling back to standard API routes.
- Quantization: The Flash variant is typically served using FP8 or INT8 quantization to maximize tokens-per-second (TPS) on standard GPU clusters.
Future ImplicationsAI analysis grounded in cited sources
Timeline
- 2024-01DeepSeek releases initial V3 series models.
- 2025-05Vercel launches AI Gateway to unify model provider access.
- 2026-07DeepSeek V4 Flash (0731) is officially released.
Weekly AI Recap
Read this week's curated digest of top AI events →
AI-curated news aggregator. All content rights belong to original publishers.
Original source: Vercel News ↗
This is a summary, not the original. Read the source, or get the weekly briefing.
The weekly digest
One email a week. Unsubscribe anytime.