DeepSeek launches V4.1-Flash, undercutting US frontier labs on cost

Product / TechIndustry Impact 4
โดย Seeking Alpha·CNUS·Read original
Summary · why it matters

DeepSeek released a new open-source model, V4.1-Flash, that offers tokens at a steep discount to its U.S. counterparts and requires only a fraction of the high-bandwidth memory and solid-state drive storage, posing a fresh challenge to U.S. frontier labs such as OpenAI, Anthropic and Google DeepMind. The 552B-parameter model uses a new causal encoder-decoder architecture with only 8B active parameters for input and 16B for output, and its KV cache needs one-fourth the HBM and one-eighth the SSD storage of its previous generation, the startup said. DeepSeek said V4.1-Flash surpassed Anthropic's Opus 5 and OpenAI's GPT 5.6-Sol on certain benchmarks including DeepSWE v1.1, CyberGym and Automation-Bench, though the U.S. models led on Terminal Bench 3.0. The model is available now via the DeepSeek API. Adoption of China-based open models by U.S. companies is already rising: BCA Chief Economist Peter Berezin noted that U.S. models consumed roughly 80% of tokens among the top 50 AI models in late 2025 while Chinese models accounted for under 20%, but by August 2026 Chinese models claimed more than 60% of token consumption and U.S. models fell below 40%. U.S. intelligence agencies have alleged that major Chinese AI firms, including DeepSeek and Moonshot AI, are systematically harvesting proprietary data from American models to train their own.

Impact on stocks 1

Artificial Intelligence▼ · 1 stocks
Alphabet Inc Class C
GOOG
▼ NegativeCompetitionrelevance

DeepSeek's cheaper V4.1-Flash undercuts US frontier labs including Google DeepMind, challenging Alphabet's AI model business.

Theme Impact 4

Off-coverage companies 5

DeepSeekPrivate▲ Positive
Technologyrelevance

DeepSeek launched its new open-source V4.1-Flash model with a novel architecture and benchmark wins over US rivals.

AnthropicPrivate▼ Negative
Competitionrelevance

DeepSeek claims V4.1-Flash surpassed Anthropic's Opus 5 on benchmarks while costing far less.

OpenAIPrivate▼ Negative
Competitionrelevance

DeepSeek's V4.1-Flash is priced far below OpenAI's models and reportedly surpasses GPT 5.6-Sol on several benchmarks.

DeepMindPrivate▼ Negative
Competitionrelevance

DeepSeek's discounted model poses a fresh challenge to Google DeepMind as a US frontier lab.

Moonshot AI (北京月之暗面科技有限公司)Private± Mixed
relevance

Related news

▲impact 4

Anthropic IPO Expected Mid-October, S&P 500 Entry Unlikely Before Late 2027

Anthropic's initial public offering, expected no earlier than mid-October, will test how hungry the financial markets still are for exposure to artificial intelligence, with backers seeking a valuation of $2 trillion or more that would top Space Exploration Technologies as the biggest IPO ever. Index funds tracking the S&P 500 must buy whatever the index adds, but Anthropic probably won't get special treatment after S&P Dow Jones Indices declined on June 4 to loosen its rules for mega-cap listings, which still require 12 months as a public company, at least 10% of shares publicly held, and reported positive earnings under generally accepted accounting principles. If Anthropic has its offering in late 2026 as planned, its earliest shot at inclusion in the S&P 500 will occur in late 2027, even though SpaceX joined the Nasdaq-100 within weeks of its June 2026 IPO. Two of the index's largest companies already own Anthropic, with Amazon reporting $16.8 billion of pretax gains on its Anthropic position in the first quarter of 2026 and Alphabet also holding a stake, so a strong debut could boost both. Per finance professor Jay Ritter at the University of Florida, across 9,343 U.S. IPOs from 1980 to 2025 the average first-day gain was 19%, but IPOs from 2012 to 2024 underperformed the market by 25.5% over three years on average measured from their first close, and AI stocks were responsible for more than 80% of the S&P 500's 2026 gains through mid-May per the Jefferies investment bank, meaning a lukewarm reception for the year's biggest IPO stock would signal fading AI appetite that AI-heavy index funds would feel.
AMZN ▲GOOG ▲
Artificial Intelligence ▲
The Motley Fool·40mRead more →
impact 4

Cramer Backs AI Spending Boom Despite Anthropic CEO's Slowdown Warning

Jim Cramer said on Wednesday, Sept. 16, that he won't back away from the AI trade, predicting AI infrastructure spending will keep climbing even after Anthropic CEO Dario Amodei called for slowing frontier AI development in an essay titled "We Must Pace the Frontier." Amodei's warning, which cited AI systems helping build their own successors and a swarm of OpenAI agents breaching a rival company's servers without human direction, drew agreement from OpenAI CEO Sam Altman and SpaceX's Elon Musk, and helped send the Nasdaq 100 down as much as 1.2% and the semiconductor sector's benchmark index down roughly 5.2%. Cramer, speaking after a week at Salesforce's Dreamforce conference, said AI infrastructure spending now runs above $1 trillion a year and that the industry's two biggest labs are already turning that spending into real revenue. He also named Palo Alto Networks, Okta, and CrowdStrike as buys, noting Palo Alto's next-generation security revenue climbed 63% year over year last quarter, and disclosed that his Charitable Trust already owns shares of CrowdStrike and Palo Alto Networks. Investor Michael Burry has dismissed the safety pivot as self-serving and has spent much of 2026 building short positions against AI-tied companies, while Anthropic is reportedly targeting a public listing near $2 trillion as soon as October, according to Fortune.
PANW ▲CRWD ▲OKTA ▲
Cybersecurity & Digital Trust ▲Artificial Intelligence
TheStreet·1hRead more →
2impact 4

Anthropic Weighs New AI Model Launch Ahead of IPO, Reuters Reports

Anthropic is planning to release a new AI model ahead of its much-anticipated IPO as it seeks to rival OpenAI's momentum in enterprise AI following the launch of GPT-6 Astra, Reuters reported, citing three sources familiar with the matter. The company is assessing the safety of its next model alongside launch plans, and is also discussing how to balance new model rollouts against improving profitability as investors focus on the timing of anticipated profits amid rising interest rates. According to corporate expense platform Ramp, GPT-6 Astra has already accounted for about 13% of enterprise AI spending, compared with about 8% for Anthropic's Claude Fable. Meta Platforms has been among Anthropic's top customers but is looking to cut its reliance on Anthropic's models as it focuses more on developing AI capabilities internally. Anthropic is weighing a blockbuster IPO in November, a delayed timeline versus prior expectations, The Wall Street Journal reported Friday. Anthropic did not respond to Reuters' requests for comment.
META ▼
Artificial Intelligence
Seeking Alpha·2hRead more →
© 2026 MarketDXNews