AI2 Releases Fully Open-Source Web Agent MolmoWeb: Controlling Web Pages Using Only "Vision"

marsbit2026-03-26 tarihinde yayınlandı2026-03-26 tarihinde güncellendi

Özet

AI2 has released MolmoWeb, a groundbreaking, fully open-source web agent that operates solely by analyzing screenshots, marking a significant leap in vision-driven web navigation. Unlike traditional agents that rely on DOM, MolmoWeb captures and interprets visual data to make decisions—such as clicking, scrolling, or typing—making its process transparent and robust. Despite its compact size (4B and 8B parameters), MolmoWeb performs impressively: it scores 78.2% on the WebVoyager benchmark, nearing OpenAI’s proprietary o3 model (79.3%), and achieves up to 94.7% success with multiple attempts. It even surpasses Anthropic’s Claude3.7 in UI element localization. AI2 also released MolmoWebMix, a massive open dataset with 36K human-browsing tasks, over 2.2M screenshot-QA pairs, and GPT-4o-verified synthetic data. The model and data are fully available on Hugging Face and GitHub under Apache 2.0, promoting transparency and collaboration in AI development. Challenges remain in complex instructions, logins, and legal compliance.

The Allen Institute for Artificial Intelligence (AI2) recently released the groundbreaking fully open-source web agent MolmoWeb . Unlike traditional agents that rely on a webpage's underlying code (DOM), MolmoWeb makes decisions solely by reading screenshots, marking a significant leap forward in "vision-driven" web navigation technology.

Core Technology: "Seeing" Web Pages Like a Human

MolmoWeb's operating logic is very intuitive: it captures a screenshot of the current browser window, decides the next action (such as clicking, scrolling, or paging) through visual analysis, then executes it and repeats. This "what you see is what you get" model makes it more robust than traditional agents because the visual layout of a webpage is generally more stable than its underlying code, and its decision-making process is completely transparent and explainable to human users.

Performance Leap: Small Model Outperforms Giants

Despite having parameter sizes of only 4B and 8B, MolmoWeb demonstrates a "small but mighty" performance:

Topping the Charts: In the WebVoyager test, the 8B version scored an impressive 78.2%, not only ranking among the top open-source models but also approaching the performance of OpenAI's proprietary model o3 (79.3%).
Huge Potential: Research found that by running tasks multiple times and selecting the optimal result, its success rate could further jump to 94.7%.
Precise Localization: In UI element localization benchmark tests, it even surpassed Anthropic's Claude3.7.

Data Support: The Largest Open Dataset to Date

AI2 has not only open-sourced the model weights but also contributed a massive dataset named MolmoWebMix. This dataset contains:

36,000 real browsing tasks completed by human volunteers.
Over 2.2 million screenshot-question-answer pairs.
Automated synthetic data verified by GPT-4o. Experiments show that synthetic data is even better than human trajectories at guiding the agent to find the "optimal path".

Open-Source Spirit and Future Challenges

Currently, MolmoWeb is fully available under the Apache 2.0 license on Hugging Face and GitHub. Although it still faces challenges in handling complex instructions, login authentication, and legal compliance (such as terms of service), AI2 firmly believes that only through complete transparency and community collaboration can we truly counter the data monopoly of large tech companies.

İlgili Sorular

QWhat is the name of the fully open-source web agent released by the Allen Institute for AI (AI2) that navigates using only screenshots?

AThe web agent is called MolmoWeb.

QHow does MolmoWeb's approach to web navigation differ from traditional web agents?

AUnlike traditional agents that rely on a webpage's underlying code (DOM), MolmoWeb makes decisions by reading and analyzing screenshots, making it a 'vision-driven' technology.

QWhat was the performance score of the 8B parameter version of MolmoWeb on the WebVoyager test, and how does it compare to OpenAI's model?

AThe 8B version scored 78.2% on the WebVoyager test, which is very close to the performance of OpenAI's proprietary model o3, which scored 79.3%.

QWhat is the name of the large, open dataset released alongside MolmoWeb, and what does it contain?

AThe dataset is called MolmoWebMix. It contains 36,000 real browsing tasks completed by human volunteers, over 2.2 million screenshot-QA pairs, and automated synthetic data verified by GPT-4o.

QOn which platforms has MolmoWeb been made available, and under what license?

AMolmoWeb has been fully released on Hugging Face and GitHub under the Apache 2.0 license.

İlgili Okumalar

Analyzing the Impact of AI on Economic Growth and Productivity

**Title: Analyzing AI's Impact on Economic Growth and Productivity** This article examines three contrasting views on AI's influence on economic growth and productivity. **The Optimistic View** posits that AI, especially through automating R&D ("recursive self-improvement"), could dramatically accelerate growth, even triggering a technological "singularity" with explosive, potentially infinite, economic expansion. **The Moderate/Mainstream View** acknowledges AI's productivity benefits but emphasizes significant real-world constraints that could limit its impact. These include: limited cost savings per task, structural ceilings on which jobs and industries are "exposed" to AI, adoption bottlenecks (e.g., compute, energy, regulatory hurdles), and the "weak link" effect where non-automatable tasks cap overall gains. Consequently, the realized AI dividend may be far lower than optimistic projections, with estimates typically ranging from 0.1% to 1.3% annual productivity growth. **The Pessimistic View** stems from two strands. The first aligns with the moderate view but applies extremely conservative assumptions about task exposure and efficiency gains, yielding minimal projected impact. The second introduces a demand-side critique: if AI primarily replaces rather than augments labor, it could depress labor's share of income, weaken consumer demand, and create a "demand trap" that ultimately stifles growth, unless offset by redistribution policies. **The authors' assessment** is nuanced: * **Short-term (1-2 years):** AI will support growth primarily through investment spending, not significant productivity gains. * **Medium-term (3-5 years):** Three potential paths emerge based on AI demand and bottleneck severity: 1. **"Optimistic Path":** High demand, few bottlenecks. Rapid productivity gains but risk of major job displacement and social conflict without redistribution. 2. **"Moderate Path" (most likely):** High demand but significant, surmountable bottlenecks. Leads to moderate productivity gains, financial market volatility (K-shaped returns), and sectoral job losses. 3. **"Pessimistic Path":** Low demand or severe bottlenecks. Minimal productivity and growth impact, triggering financial market corrections but allowing a smoother societal transition with less labor disruption. * **Long-term:** AI holds potential for a major productivity revolution and prosperity. The conclusion stresses that no path is smooth. Technologically "optimistic" outcomes could be socially detrimental, while "pessimistic" technological diffusion might be more socially stable. Policymakers must monitor developments and prepare balanced responses to manage economic, financial, and social sustainability.

marsbit3 dk önce

Analyzing the Impact of AI on Economic Growth and Productivity

marsbit3 dk önce

IRS Fraud Alert: Fake Letters Target Crypto Holders in Data-Stealing Scam

The IRS Criminal Investigation unit has warned cryptocurrency holders of a sophisticated phishing scam involving fake paper letters that mimic official IRS correspondence. These letters, which include official formatting and a deadline, instruct recipients to enroll in a non-existent "Digital Asset Compliance Portal." A QR code inside the letter directs victims to a spoofed IRS website designed to steal personal information and, in some cases, trick them into directly transferring cryptocurrency to scammers. The scam exploits the fact that the IRS has legitimately sent compliance letters regarding digital assets since 2019, making such mail seem plausible. The agency clarifies it never uses QR codes in official letters, never asks for digital asset transfers as part of compliance, and does not operate a digital asset enrollment portal. Legitimate notices can be verified via IRS.gov. If you receive such a letter, do not scan the QR code, call any listed numbers, or enter personal information on any linked site. Verify your account status only through official IRS tools. This scam undermines trust in genuine IRS communications during a period of increased crypto enforcement.

TheNewsCrypto7 dk önce

IRS Fraud Alert: Fake Letters Target Crypto Holders in Data-Stealing Scam

TheNewsCrypto7 dk önce

The New Cold War is a Tech Stock War

The New Cold War is a Tech Stock War The article argues that the contemporary geopolitical and economic rivalry between the US and China represents a "New Cold War," but one fundamentally fought through technology and financial markets, not physical barriers or conventional trade. Historically, US dominance was secured through financial systems. The Soviet Union, reliant on the rigid "Transferable Ruble," was ultimately undermined by its dependency on the US dollar for oil trade. Later, Japan's semiconductor challenge was countered not just by tariffs (e.g., Plaza Accord, 301 investigations) but by binding it to US Treasury bonds. China presents a more complex, "embedded" challenger. While it holds vast dollar reserves and US debt like Japan, its industrial base is stronger and more diversified than the Soviet Union's. Surviving the initial 2018 trade war phase, the conflict has evolved into a "tech-financial war." The core battlefield is now the stock market. US tech stocks (AI, semiconductors) are treated as sovereign assets, buoyed by bipartisan national will. China is pushing to strengthen its own financial markets to convert industrial strength into financial power and fund its tech ambitions. Companies like ChangXin (semiconductors), Moonshot AI, and DJI compete not just for market share but as financial proxies for their respective systems. The new paradigm is moving from globally efficient monopolies (Apple, Google) towards companies that achieve monopolistic profits within their respective geopolitical spheres. This competition over "pricing power" and financial valuation in segmented markets defines the current era, making the stock market the primary arena for this tech-centric struggle.

marsbit12 dk önce

RWA Weekly: Ten European Financial Institutions Establish Tokenized Asset Cooperative; Ondo Launches New Execution Network Ondo Network

RWA Weekly: European Banks Form Tokenized Asset Cooperative; Ondo Launches New Execution Network Ondo Network Covering July 24-31, 2026, the RWA sector saw a steady on-chain total value locked (TVL) of $36.8 billion, with holder count hitting a record high. However, stablecoin transfer volumes fell sharply (~30%), indicating low on-chain settlement demand. Key regulatory moves include South Korea advancing stablecoin legislation and a push to scrap crypto taxes, Kenya lowering capital requirements for stablecoin issuers, and Zimbabwe approving seven projects for its crypto sandbox. In project developments, BIS-led Project Agorá successfully tested cross-border payments with tokenized funds across six currencies. Ten major European financial institutions formed the RL1 blockchain cooperative to build tokenized asset infrastructure. Other notable updates: Aviva launched a tokenized dollar liquidity fund on XRPL, POSCO International tokenized commercial invoices on Injective, and a Brazilian farmer used tokenized cattle as collateral for a loan. Additional progress includes BNY Mellon migrating its core transfer agent operations to blockchain, Securitize gaining SEC investment advisor registration, and Tether’s compliant stablecoin USA₮ launching on Celo. Ondo Finance introduced Ondo Network, a new execution layer focused on speed and privacy, moving away from its initial chain plans. An analysis highlights that despite the growing scale of on-chain RWAs (~$32B), approximately 90% remain underutilized in DeFi, pointing to a critical challenge in unlocking liquidity and fostering real-world application beyond mere issuance.

marsbit13 dk önce

RWA Weekly: Ten European Financial Institutions Establish Tokenized Asset Cooperative; Ondo Launches New Execution Network Ondo Network

marsbit13 dk önce

South Korean Stock Market Sees Sharp Rebound After Forceful De-leveraging, SK Hynix Rises 30%

On July 31, South Korean stocks staged a historic rebound. The benchmark KOSPI index surged 18.27%, with chipmaker SK Hynix hitting a 30% gain limit. This followed a brutal, near-40% decline in the KOSPI over the previous month, driven largely by a deleveraging spiral involving leveraged ETFs. Analysts attributed the sharp sell-off to structural liquidity issues rather than deteriorating corporate fundamentals. The rally was triggered by a confluence of positive catalysts. Firstly, strong earnings from U.S. cloud giants Microsoft and Amazon alleviated fears of an "AI bubble burst," boosting global tech sentiment. Secondly, SK Group Chairman Chey Tae-won made a rare personal purchase of SK Hynix shares, seen as a strong vote of confidence. Thirdly, the South Korean government announced a 20 trillion won ($139 billion) AI investment fund. In response to the market turmoil, South Korean regulators are tightening controls on leveraged ETFs, admitting oversight shortcomings. Measures include raising minimum cash保证金 requirements for散户 investors and suspending new product launches. While the rebound signals eased liquidity pressure, analysts note deep structural issues remain. The market's future stability is seen as dependent on global tech capital expenditure trends and memory chip price cycles, with some viewing the surge as a technical correction rather than a definitive trend reversal.

marsbit33 dk önce

South Korean Stock Market Sees Sharp Rebound After Forceful De-leveraging, SK Hynix Rises 30%

marsbit33 dk önce

İşlemler

Spot