DeepSeek Unveils Full Multimodal Support in Major Update, Hints at V4 Visual Version
Less than a week after launch, the open-source AI agent framework DeepSeek Harness has rolled out a major update, officially adding robust multimodal capabilities. This release enables the DeepSeek model adapter to handle native image inputs, fueling industry speculation that a dedicated vision model from DeepSeek is imminent.
Under the hood, this update delivers tangible UX enhancements. Commands like /goal and /plan, previously limited to text, now accept image inputs alongside text. The @ mention feature has also expanded beyond file extensions to include references to past conversations, allowing developers to combine screenshots, local files, and chat history in a single prompt. To stabilize complex workflows, the update ensures persistent image attachments for both MCP and ACP protocols, preventing data loss after a single call. It also resolves bugs related to oversized images and payload bloat from accumulated history in multi-turn chats.

Beyond multimodal upgrades, the release restructures component distribution and architecture. Claude Code and Codex are no longer bundled with the core product; instead, they are available as optional Profile Bundles. Codex introduces two key features: a non-interactive permission mode for unattended processes and support for multiple named instances, allowing distinct sub-agents to run concurrently. Additionally, the sub-agent reporting path is optimized, enabling reportDelivery to instantly notify and wake parent tasks, eliminating the need for blind waiting.

Other improvements include concurrent query support for web_search and persistent PowerShell session support in the Windows PTY terminal, now enabled by default in minimalist mode. SQLite backend performance—covering read/write and fork operations—has been significantly boosted, with reduced storage overhead. Finally, the official team confirms that "DeepSeek Harness" is now a registered trademark, with updated brand usage guidelines.
Related article
Suno to Watermark Songs Amid Legal Battles
Suno, the platform enabling users to generate AI-created music, has unveiled new features to label platform-produced tracks, restrict downloads, and update community standards to curb unauthorized replicas. These updates arrive as Suno confronts mult
Musk Admits Grok Build Leaked User Code, Promises to Erase All Historical Data
Elon Musk directly addressed the privacy controversy surrounding Grok Build, beginning with a simple "True" to confirm the incident's validity. He pledged that all user data previously uploaded to SpaceXAI would be permanently erased, stating, "not a
U.S. Stocks Hit Historic Milestone as AI and Aerospace Giants Prepare for Trillion-Dollar Debut
Elon Musk, Sam Altman, and Dario Amodei, three titans of the technology sector, are advancing toward initial public offerings for their respective ventures. With SpaceX, OpenAI, and Anthropic—three industry behemoths nearing trillion-dollar valuation
Related Special Topic Recommendations
Comments (0)
0/500
Less than a week after launch, the open-source AI agent framework DeepSeek Harness has rolled out a major update, officially adding robust multimodal capabilities. This release enables the DeepSeek model adapter to handle native image inputs, fueling industry speculation that a dedicated vision model from DeepSeek is imminent.
Under the hood, this update delivers tangible UX enhancements. Commands like /goal and /plan, previously limited to text, now accept image inputs alongside text. The @ mention feature has also expanded beyond file extensions to include references to past conversations, allowing developers to combine screenshots, local files, and chat history in a single prompt. To stabilize complex workflows, the update ensures persistent image attachments for both MCP and ACP protocols, preventing data loss after a single call. It also resolves bugs related to oversized images and payload bloat from accumulated history in multi-turn chats.

Beyond multimodal upgrades, the release restructures component distribution and architecture. Claude Code and Codex are no longer bundled with the core product; instead, they are available as optional Profile Bundles. Codex introduces two key features: a non-interactive permission mode for unattended processes and support for multiple named instances, allowing distinct sub-agents to run concurrently. Additionally, the sub-agent reporting path is optimized, enabling reportDelivery to instantly notify and wake parent tasks, eliminating the need for blind waiting.

Other improvements include concurrent query support for web_search and persistent PowerShell session support in the Windows PTY terminal, now enabled by default in minimalist mode. SQLite backend performance—covering read/write and fork operations—has been significantly boosted, with reduced storage overhead. Finally, the official team confirms that "DeepSeek Harness" is now a registered trademark, with updated brand usage guidelines.
Suno to Watermark Songs Amid Legal Battles
Suno, the platform enabling users to generate AI-created music, has unveiled new features to label platform-produced tracks, restrict downloads, and update community standards to curb unauthorized replicas. These updates arrive as Suno confronts mult
Musk Admits Grok Build Leaked User Code, Promises to Erase All Historical Data
Elon Musk directly addressed the privacy controversy surrounding Grok Build, beginning with a simple "True" to confirm the incident's validity. He pledged that all user data previously uploaded to SpaceXAI would be permanently erased, stating, "not a
U.S. Stocks Hit Historic Milestone as AI and Aerospace Giants Prepare for Trillion-Dollar Debut
Elon Musk, Sam Altman, and Dario Amodei, three titans of the technology sector, are advancing toward initial public offerings for their respective ventures. With SpaceX, OpenAI, and Anthropic—three industry behemoths nearing trillion-dollar valuation





Home






