DeepSeek's multimodal model nears Anthropic's Opus 4.8 at 99% lower cost

DeepSeek's experimental V4-Flash-Vision-Exp model matches its flagship text model on reasoning while lifting multimodal agent performance near Anthropic's Opus 4.8, intensifying a price war that threatens the premium pricing US AI leaders have relied on to justify heavy investment.

"Chinese labs have essentially caught up with US leaders on capability while pricing output at a fraction of the cost," Confluence Investment Management said in a note to clients.

The Hangzhou-based startup released the model Friday on its API platform, adding visual understanding to the text-only V4 Flash. The model can analyze images and screenshots, and DeepSeek said it performs close to Opus 4.8 on multimodal agent benchmarks. DeepSeek priced V4 Flash at about $0.28 for output that costs $25 with Anthropic's Opus 4.8.

The gap threatens Anthropic's premium positioning ahead of a planned IPO that the company expects to match or top SpaceX's record size. Anthropic's annualized revenue has surpassed $65 billion.

The price gap widens

DeepSeek's aggressive pricing follows a broader repricing across the sector. OpenAI cut the price of its GPT-5.6 Luna model by 80 percent from its launch price three weeks earlier, bringing it to $1.20 for the same output that costs $0.28 with V4 Flash and $25 with Opus 4.8. Moonshot AI's Kimi K3, released days before V4 Flash, rivals both Opus 4.8 and OpenAI's GPT-5.6 Sol on benchmark tests.

The pattern echoes China's playbook in steel, rare-earth processing, automobiles and electronics, where subsidized producers undercut foreign rivals to capture market share. Beijing has made AI a priority industry, and analysts expect the same pricing strategy to apply. DeepSeek did not disclose the test conditions for its multimodal benchmark comparison, leaving the Opus 4.8 gap unverified against independent results.

What it means for investors

The competitive pressure lands as US firms pour billions into model development and data-center infrastructure. Analysts estimate AI investment accounts for roughly one-third of current US economic growth, with spending lifting demand for everything from microprocessors to cooling equipment. Nvidia, whose GPUs power much of the buildout, has been the biggest beneficiary.

If Chinese models sustain near-parity performance at a fraction of the cost, the economics of the US AI buildout come into question. Anthropic is trying to position its models as a premium product, but DeepSeek's pricing shows how hard that defense may be to hold. Investors are watching whether the price war forces US firms to cut prices, compress margins, or slow infrastructure spending, and whether the repricing spreads to the chipmakers and data-center operators that have ridden the AI boom.

This article is for informational purposes only and does not constitute investment advice.