According to DeepSeek’s technical report, on April 30, the company introduced Visual Primitives, a method that embeds basic visual units such as points and bounding boxes into reasoning chains to address the Reference Gap problem in multimodal tasks. The method reduces image token consumption through KV cache compression.
In counting and spatial reasoning benchmarks, the approach matches the performance of GPT-5.4, Claude-Sonnet-4.6, and Gemini-3-Flash across select dimensions. DeepSeek stated it will open-source portions of the benchmarks and datasets, with model weights to be released following integration.
Disclaimer: The information on this page may come from third parties and does not represent the views or opinions of Gate. The content displayed on this page is for reference only and does not constitute any financial, investment, or legal advice. Gate does not guarantee the accuracy or completeness of the information and shall not be liable for any losses arising from the use of this information. Virtual asset investments carry high risks and are subject to significant price volatility. You may lose all of your invested principal. Please fully understand the relevant risks and make prudent decisions based on your own financial situation and risk tolerance. For details, please refer to
Disclaimer.
Related Articles
Attackers Hijack TanStack, OpenSearch, Mistral Official Pipelines, Push 84 Malicious Versions on May 12
According to Beating's monitoring, on May 12 at 3:20–3:26 UTC+8, attackers affiliated with TeamPCP hijacked the official release pipelines of TanStack, Amazon's OpenSearch, and Mistral, pushing 84 malicious package versions across npm and PyPI. Affected packages include @tanstack/react-router (10M+
GateNews5m ago
Ixirpad Partners With Cware Labs to Support AI and Web3 Startups
According to an announcement on May 11, Ixirpad entered into a strategic partnership with Cware Labs to accelerate sustainable infrastructure development in the Web3 industry. Cware Labs, operating as a venture studio, will identify and support high-potential blockchain and AI projects. The
GateNews19m ago
Claude Code Agent View: single-screen management concurrent sessions
Anthropic on May 11 introduced a new feature for Claude Code called “Agent View,” which consolidates multiple simultaneously running Claude Code work sessions into a single screen for management, eliminating the need to switch back and forth between multiple terminal tabs. According to Anthropic’s official blog, the feature is rolling out in a Research Preview format and is available for Pro, Max, Team, Enterprise, and Claude API solutions. A single post on the official X account has received mo
ChainNewsAbmedia27m ago
Karpathy Endorses HTML Output for Large Language Models, Predicts Interactive Neural Video as Ultimate Form
According to Andrej Karpathy, OpenAI founding member and "vibe coding" concept creator, today he endorsed the Claude Code team's approach of using HTML instead of Markdown for large language model outputs. Karpathy outlined an evolution roadmap for AI interaction interfaces: from plain text to Markd
GateNews54m ago
Austrac Warns of AI-Driven Money Laundering Risks as Australia Expands Anti-Money Laundering Rules July 1
According to Austrac, on May 12, Australia's financial intelligence agency warned that artificial intelligence is raising money laundering risks by enabling criminals to fabricate identities, forge documents, and hide proceeds faster and at greater scale.
Starting July 1, 2026, real estate
GateNews59m ago
Google: Large language models are being used for real-world attacks; AI can bypass dual-factor authentication security mechanisms
According to CoinEdition on May 12, Google’s Threat Intelligence Group released a report warning that attackers have used large language models in real-world cyberattacks affecting global systems, and confirmed that hackers have developed a Python-based zero-day vulnerability that can bypass two-factor authentication (2FA) security mechanisms; Google said that related activity is linked to state-sponsored cyberattacks and the abuse of AI tools within underground hacker networks. Specific Applica
MarketWhisper1h ago
