DeepSeek V4 is here: How it compares to ChatGPT, Claude, Gemini

DeepSeek released a preview of DeepSeek V4, an open-source AI model available under an MIT license with two versions: V4-Pro at 1.6 trillion parameters and V4-Flash at 284 billion total parameters. The model is positioned as far cheaper than competing frontier systems, with pricing of $1.74 per 1 million input tokens and $3.48 per 1 million output tokens versus $5/$30 for GPT-5.5, $5/$25 for Claude Opus 4.7, and $2/$12 for Gemini 3.1 Pro. The article frames the launch as a competitive move in the U.S.-China AI race, though current leaderboard performance still trails the top frontier models.

Analysis

DeepSeek’s real significance is not that it narrows the frontier-model gap; it is that it compresses pricing power across the entire model stack. If a credible, low-cost open model can handle a large share of coding and agentic workflows, the margin pool shifts away from model vendors toward the layers that own distribution, workflow integration, and proprietary data. That is structurally negative for closed-model monetization assumptions and positive for any platform that can swap model providers without customer-visible degradation.

The second-order effect is a faster commoditization cycle for inference, which should pressure cloud attach rates per task even if total token volumes keep rising. In the near term, this is bearish for premium AI API pricing and likely forces competitors into either price cuts or bundling, but it is bullish for compute demand at the application layer because cheaper tokens expand use cases and raise throughput. The winner set broadens to orchestration, evaluation, and enterprise software vendors that sit above the model layer and can arbitrage price/performance across providers.

The biggest market miss is that lower model prices are not automatically bearish for all AI infrastructure; they can be bearish for model gross margins while still bullish for aggregate GPU consumption if usage explodes. The true risk to that bull case is a sudden acceptance that open-source models are “good enough,” which would shift spending from frontier model subscriptions to self-hosted deployments over the next 6-18 months. That would matter most for vendors with the least differentiated distribution and the most exposed pricing power, while accelerating geopolitical bifurcation in enterprise AI procurement.

Key Decisions for Investors

• Short basket: premium closed-model monetizers vs. open-model enablers over 1-3 months; express via short ARKK-style AI beta on names most dependent on proprietary API pricing, funded with longs in enterprise software/platform vendors that can route between models.
• Long semis / compute, but hedge model-price compression: pair long SMH or NVDA against short exposure to AI software names with weak moat and high gross-margin expectations; use a 3-6 month horizon and take profits if model pricing wars intensify.
• Accumulate beneficiaries of model-switching and workflow orchestration over 6-12 months; prioritize platform/software names with multi-model abstraction layers and low customer concentration, as they gain bargaining power when frontier APIs commoditize.
• Avoid chasing pure-play frontier model enthusiasm into earnings; the asymmetry is worse for vendors priced for durable premium API monetization. Use put spreads or call overwrites where implied volatility still embeds strong pricing power assumptions.
• Watch for a reopen in enterprise open-source deployment demand over the next 1-2 quarters; if adoption data confirms “good enough” performance, rotate from model vendors into cloud-neutral infrastructure and integration plays.