DualEntry's Accounting AI Benchmark (DualBench100) finds DeepSeek V4.1 Flash 1017 correctly completed 82.2% of tasks, down from 84.2% for the earlier V4.1 Flash, tying for tenth of 19 models. Open models are close on accuracy, with GLM-5.3 Flash leading them at 88.1%, while Gemini 4 Argon tops the chart at 91.1%. The bigger gap is consistency: Gemini 4 Argon passed 86.1% of tasks in all three runs, versus 76.2% for GLM-5.3 Flash. The benchmark covers 101 synthetic accounting cases, with each model run three times and an 80% pass mark. DualEntry founder Benedict Dohmen said reliability, not accuracy, is now the key issue for agents working in the general ledger.
Summary written by Upside AI from the original article. Read the source for full details.
Read the original on Yahoo Finance ↗
Keywords
DeepSeekartificial intelligenceChina
Open a keyword to see other news on the same topic.
Related tickers and likely impact
GOOGL
• NeutralAlphabet Inc. (Class A)
Relevance: low
| Market | US stocks |
|---|---|
| Sector | Communication Services |
| Industry | Interactive Media & Services |
| Market cap | $4.3T |
| Close on 10/20/2026 | 340.28 |
| Change that session | -1.13% |
Impact is scored by AI from the headline and may be wrong. It is not investment advice.
Related news
- DeepSeek Completes Nearly $15 Billion Funding Round, Largest Ever in China
Chinese AI startup DeepSeek is finalizing a funding round of about 100 billion yuan (US$14.98 billion) from existing and new investors, including Tencent and CATL, according to Reuters citing sources. The round values the company at 500 billion yuan (about $74.9 billion) before the new capital, and its size is double the initial target of 50 billion yuan. Other participants include Midea, Meituan, Geely and GigaDevice, with more than 15 investors in total. The terms and investor list are not final and the deal is expected to close within days, while DeepSeek, CATL and Tencent have not commented. The move follows DeepSeek's hiring of CITIC Securities in September to prepare a potential IPO on China's STAR Market.
- U.S. bans DeepSeek from government agency systems amid AI rift with China
The U.S. Department of Homeland Security made public an order requiring federal civilian agencies to remove DeepSeek-related technology from their IT systems and barring them from buying DeepSeek products and services, including third-party products built on its technology. The order was originally issued on July 24 and covers agencies such as the Treasury, State and Justice departments but excludes the Department of Defense. DHS warned that DeepSeek's models pose multiple threats, including possible transfer of sensitive government data to China and critical security flaws in its iOS app. Agencies may seek exceptions, including compliance delays, by submitting written justifications within 60 days of the order. The ban follows earlier restrictions, including Congress's limits in December 2025 and the Defense Information Systems Agency's block in January 2025.
- US bans DeepSeek from civilian agency IT systems
The US Department of Homeland Security has ordered the removal of DeepSeek products from federal civilian agency IT systems and barred agencies from new purchases, with the order also covering DeepSeek affiliates and third-party products built on DeepSeek. DHS warned that use of DeepSeek by employees and contractors could expose sensitive information to China, including details about US citizens and federal programs, and cited security flaws in DeepSeek's iOS app. The order took effect immediately upon publication and expands earlier bans at the Commerce Department and parts of the defense, security and intelligence community, amid US-China competition in AI. Australia had already banned DeepSeek products from government systems and devices in February 2025.
- US Orders Civilian Agencies to Rip DeepSeek Out of Government Systems
The U.S. Department of Homeland Security on Monday ordered civilian federal agencies to remove DeepSeek products and third-party services built on the model from their IT systems, and to stop buying them. The notice warns that using DeepSeek may transfer sensitive data to China and could produce intelligence on U.S. persons and agencies. The Commerce Department, the Navy, and the states of New York, Tennessee and Virginia had already restricted DeepSeek. The FCC votes on October 29 on barring Chinese labs from testing electronics for U.S. use.
- One Reason Alphabet Stock Looks Better Than Its Price
Alphabet (GOOGL) shares fell 1.2% over the six months to October 21, 2026, while the S&P 500 gained 9.3%, even as Google Cloud grew quickly: second-quarter fiscal 2026 revenue reached $24.8 billion, up 82%, and operating margin rose to 35.6% from 20.7%. Cloud's backlog reached $514 billion at the end of June 2026. However, Alphabet spent $44.9 billion on capital projects in the quarter, more than the $39.1 billion of operating cash flow, and raised its 2026 capex guidance to $195–205 billion from $180–190 billion. The main risks cited are limited capacity and costs rising faster than backlog converts to revenue, with most revenue from TPU system deals expected in 2027. Alphabet's third-quarter report is expected on October 28, 2026.
- Meta Falls 3% as Higher AI Spending Outlook Overshadows Earnings Beat; Alphabet Dips, Microsoft Holds Steady
Meta reported Q3 2026 revenue of $64.6 billion, up 26.1% year over year and above estimates, and guided Q4 revenue to $73 billion to $76 billion. Excluding a large legal settlement charge, diluted EPS was $7.40. However, Meta narrowed its full-year capex outlook to $135 billion to $145 billion and lifted total expense guidance to $175 billion to $179 billion, pushing the stock down 3% to $677.20. Unlike Alphabet and Microsoft, which also rent cloud computing capacity to outside customers, Meta earns nearly all its revenue from advertising and must fund its data centers itself; it spent $34.9 billion on capex in the quarter, posted negative free cash flow of $2.12 billion, and saw operating margin fall to 20% from 40%. Ad impressions rose 10% and the average price per ad climbed 14%, while more than 10 million people in the U.S. and Canada set up Muse within six weeks, and management expects its cost efficiency to improve more than 10x in the next year.
Related markets
Latest news
- Did Wall Street forget about Tesla's Cybercab?
Yahoo Finance's Brian Sozzi discusses Tesla (TSLA)'s Cybercab expansion and its relevance amid a possible merger with SpaceX (SPCX). Elon pointed to 2 million cumulative unsupervised robotaxi miles by quarter-end, Austin operating hours now run 6:00 a.m. to midnight, and more Cybercabs are expected in more cities in coming months. The Austin Cybercab fleet has grown by a factor of 10 since launch about a month ago. Tesla shares fell partly because the company cut capex by $5 billion, which Sozzi attributes to Optimus robot production bottlenecks rather than the Cybercab. He argues that with merger talks being explored, investors should start thinking of SpaceX and Tesla as one company.
- Crypto for Advisors: Stablecoins are ready for prime time
CoinDesk's Crypto for Advisors newsletter cites an EY-Parthenon 2026 Blockchain Payments Survey of 521 corporate and financial institution decision-makers. It found 18% of corporates have used stablecoins in production, double the 2025 share, while 15% report adopting tokenized deposits. Some 69% of financial institutions offer or plan to launch stablecoin services within two years, up from 30% in 2025. Cross-border business payments are the leading use case, and 45% of corporates and 53% of financial institutions favor using both stablecoins and tokenized deposits. Corporates strongly prefer blockchain payments integrated into existing banking interfaces (74%), and 63% would likely adopt them if offered by their current bank or payments provider.
- Is USA Rare Earth Stock a Once-in-a-Generation Reshoring Opportunity?
The article examines whether USA Rare Earth (USAR) can help the U.S. and its allies secure critical minerals, given China's dominance in rare-earth extraction and processing. China banned rare-earth exports in October 2025 in response to U.S. tariffs and chip restrictions; the ban was later temporarily lifted, though some Chinese companies have canceled U.S. exports, and China has also banned technology exports to USA Rare Earth and MP Materials. USAR shares are down roughly 15% in 2026 and about 74% from their lifetime high last October. Excluding the recently acquired Brazil-based Serra Verde, the company is essentially pre-revenue, though U.S. government supply agreements offer some pricing floor protection. With a market cap of roughly $4 billion and a target of about $1.8 billion in EBITDA for 2030, the author sees a worthwhile risk-reward profile for risk-tolerant investors, while stressing the stock is not low-risk.
- Nebius Declines 6% Despite Fresh Citi Financing; CoreWeave Drops 5%, Cloudflare Edges Higher
Nebius (NBIS) fell 6% to $162.03 despite a new senior secured credit facility from Citi, which matures in just under a year and is meant to fund data center development and GPU purchases while the company seeks longer-term asset-backed financing. CoreWeave (CRWV) dropped 5% to $58.86, while Cloudflare (NET) rose 2% to $313.35, with the article framing the selling as concentrated among neoclouds that rely on debt. A Barclays strategist noted that AI-spending credit stress has stayed largely confined to neocloud companies. Nebius has $37.5 billion in remaining performance obligations, $8.13 billion in capital spending, and three customers each accounting for a double-digit share of quarterly sales. The article says the Citi line adds to leverage concerns, making the timing and terms of the next financing a key test.
- How agentic AI shopping changes consumer staples: P&G CFO
P&G reported strong first-quarter earnings and reiterated its full-year revenue forecast. In an interview with Yahoo Finance, CFO Andre Schulten said that people using its products remains the core of the business, but the consumer's path to purchase is changing as AI agents buy on people's behalf, so P&G is proactively working out how agent-to-agent marketing works and how to distribute brand information. He said the company is also using decades of collected data and simpler algorithms to speed up molecule discovery, automate data extraction and analysis, and make forecasting more efficient. On marketing, he said algorithms weigh credibility, so P&G plans to publish and make accessible its scientific publications, independent studies and consumer reviews, an approach he believes gives big brands an advantage.
- SpaceX Just Became an AI Company. The Numbers Prove It.
SpaceX reported third-quarter revenue up 161% year over year to $14.6 billion and net income of $4.4 billion, making it profitable. Its AI segment, now called 'super intelligence (SI),' generated $7.6 billion in revenue, up 830% year over year and 195% sequentially, making it just over half of sales and the largest revenue source. Connectivity remains the most profitable segment with $2.4 billion in operating income, versus $431 million for SI, which lost $1.26 billion last quarter. SI accounted for $16.4 billion of the company's $19.6 billion in Q3 capital expenditures, but the results should ease concerns that it is a money pit, as cloud agreements with Alphabet and Anthropic and $608 million of Cursor revenue since its August acquisition show the investments are paying off. The author notes the stock remains richly valued.
Get the briefing on Telegram or Zalo
Join the Upside Telegram channel or Zalo group for the Vietnam stocks and gold briefing at 8:00 and 19:00 (Vietnam time) on trading days, plus a weekly wrap on Saturday morning.