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

Only 153 Venture Capital Firms Invested in July: Is the Crypto VC Industry Experiencing a 'Mass Extinction'?

In July 2026, only 153 unique venture capital firms participated in disclosed crypto funding rounds, marking the lowest monthly count since November 2020. This figure represents an 87% decline from the peak of 1,177 firms in 2022. Overall, the first seven months of 2026 saw crypto projects raise approximately $11.78 billion across 481 rounds. This crypto VC contraction contrasts sharply with the broader venture capital landscape, where global VC investment reached a record $560.4 billion in H1 2026, heavily fueled by major AI company financings. This shift in capital allocation has drawn funds away from the crypto sector. Within crypto, funding is highly concentrated. Trading platforms, prediction markets, and payment sectors absorbed 53% of the total capital. While early-stage deals remain frequent, the largest sums flow to a few late-stage rounds and mergers & acquisitions, which surged to $7.23 billion in Q2 2026. The market is consolidating around top funds like a16z crypto and Dragonfly, which successfully raised new multi-billion dollar funds, while many smaller firms have retreated. Analysts describe this as a "great extinction" for crypto VCs, where capital is becoming more selective, favoring proven business models and assets over early-stage speculation. This raises the bar for project quality, funding efficiency, and viable exit paths.

marsbit2 dk önce

Only 153 Venture Capital Firms Invested in July: Is the Crypto VC Industry Experiencing a 'Mass Extinction'?

marsbit2 dk önce

Storage Chip Stocks Skyrocket: Is This a Return to a Bull Market or a Dead Cat Bounce?

The storage chip sector experienced a dramatic overnight rebound in US markets, with major stocks like SK Hynix and Micron surging over 17%. This sharp reversal is attributed to multiple factors: potential short-selling bans and market stabilization measures from South Korea, significant leverage unwinding (with leveraged ETF assets down ~70% from peaks), cooler-than-expected US inflation data easing rate hike fears, and a stellar Microsoft earnings report showing strong cloud/AI demand growth. Microsoft's results, highlighting Azure revenue crossing $100 billion, helped counter the "AI bubble burst" narrative that had pressured the sector. However, the article cautions against declaring a sustained bull market. Key risks remain, including potential further interest rate hikes by the Bank of Japan, which could pressure global carry trades, and unanswered questions about whether high industry profits can be maintained amid the supply cycle and new competition. Investors are advised to monitor whether gains hold, watch for updates on HBM/DRAM/NAND pricing and orders, and await key upcoming events like NVIDIA's earnings. The conclusion emphasizes disciplined investing—avoiding panic selling or FOMO buying—and using strategies like hedging to manage risk in this volatile environment.

marsbit11 dk önce

Storage Chip Stocks Skyrocket: Is This a Return to a Bull Market or a Dead Cat Bounce?

marsbit11 dk önce

Strategy's Loss in the Second Quarter Reaches $8.22 Billion Amid Bitcoin Decline

Strategy, the largest corporate holder of Bitcoin, reported a net loss of $8.22 billion for the second quarter. This loss was primarily driven by an $8.32 billion unrealized loss on its Bitcoin holdings due to a decline in the asset's price during the period. Despite these paper losses, the company increased its Bitcoin holdings to 843,775 BTC, a 25% growth since the start of the year. As part of a new monetization strategy, Strategy sold approximately $218.4 million worth of Bitcoin, mainly to fund dividends for preferred shareholders, with $216 million of that sold after Q2 ended. The company also built a $3.75 billion cash reserve, which it claims is sufficient to cover over two years of dividend and interest payments, aiming to insulate itself from Bitcoin's volatility while meeting obligations. Following the earnings release, Strategy's stock (MSTR) rose 4.7% in regular trading but corrected slightly after-hours. This pattern reflects how the company's accounting results are heavily tied to Bitcoin's price swings, even as its long-term strategy remains unchanged. The report indicates that Strategy is maintaining its core strategy of accumulating Bitcoin while building a financial buffer. This quarterly loss follows a recognizable pattern, with the company posting significant unrealized losses in previous quarters (e.g., $12.4 billion in Q4 2025 and ~$12.5 billion in Q1 2026) due to fair-value accounting. A key technical shift is its new monetization program, which introduces periodic selling pressure on the market, transitioning Strategy from a pure accumulator to a participant that occasionally adds supply. A critical question remains: how long can the cash reserve cover dividend obligations if a Bitcoin price downturn persists beyond two years?

cryptonews.ru22 dk önce

Strategy's Loss in the Second Quarter Reaches $8.22 Billion Amid Bitcoin Decline

cryptonews.ru22 dk önce

Will Terrorist Durov Ban Russian Officials?

Telegram founder Pavel Durov publicly reacted to being labeled a "terrorist" by Russian authorities, stating the designation came after he refused demands for mass surveillance and censorship on the platform. In a Telegram post, he highlighted that this status formally bans him from "publishing information online." Durov concluded with a statement widely circulated: Russian officials "clearly don't understand who can ban whom on the internet." This remark suggests Durov could potentially restrict official Russian government and officials' channels on Telegram, which continue to operate on the platform despite its formal blocking in Russia. The situation parallels previous, slow-moving state directives, like switching officials to domestic cars, contrasted with the current push to migrate all government communication to the Russian-made messenger MAX by 2030. However, reports indicate many officials still use Telegram via workarounds, fearing surveillance on MAX, while alternatives like BiP and KakaoTalk recently became inaccessible in Russia without a VPN. Durov has not specified any immediate actions against state channels. His statement is an initial response, with further developments depending on the authorities' reaction. The dynamic differs from 2020 when Russian regulators lifted a block on Telegram; now, Durov implies control from within the platform itself over the official accounts that persisted through that earlier blockade.

cryptonews.ru22 dk önce

Will Terrorist Durov Ban Russian Officials?

cryptonews.ru22 dk önce

DeepSeek V4 Official Version Arrives, New Capabilities Emerge, Value-for-Money King Enters the Fray

On July 31st, DeepSeek officially launched the public API beta for its DeepSeek-V4-Flash model. A key highlight is its performance on multiple Agent benchmark tests, reportedly nearing or even surpassing the level of the V4-Pro preview version from three months ago. Notably, the Flash model achieves this with significantly smaller scale (130B active parameters vs. Pro's 490B), suggesting that post-training optimization and data quality may be as crucial as raw model size. DeepSeek emphasized that the V4-Flash-0731 uses the same model architecture and size as its preview version, with improvements attributed solely to "re-trained post-training." The update also marks the official debut of DeepSeek's self-developed Agent framework, "Harness." The move signals DeepSeek's strategic push to position its cost-effective Flash model as a competitive base for Agent applications—scenarios requiring autonomous planning, tool usage, and complex task execution—where inference speed and cost are critical. By natively supporting OpenAI's Responses API format and adapting for code-generation scenarios, DeepSeek aims not just to be a cheaper alternative but to establish its own ecosystem in the Agent era. This release follows DeepSeek's record-breaking ~$50 billion fundraising round roughly two months prior, underscoring market confidence in its technology and commercialization prospects. The company is reportedly preparing for another funding round at a valuation of approximately $71 billion. The Flash model's advancement represents a step in fulfilling the high expectations that come with this valuation, setting the stage for the impending release of the V4-Pro official version and intensifying competition in the global Agent landscape.

marsbit26 dk önce

DeepSeek V4 Official Version Arrives, New Capabilities Emerge, Value-for-Money King Enters the Fray

marsbit26 dk önce

İşlemler

Spot