Home
Apple’s Local AI Returns With oMLX 0.3.9: Gemma 4 Vision Boost, One-Click Copilot, and Balanced Cloud Large Model Power
As of May 13, Beijing Time, the Apple Silicon local AI ecosystem has achieved a major milestone. The release of oMLX version 0.3.9.dev2 integrates several advanced optimization technologies, substantially boosting the speed and accessibility of local large language models for image and text tasks. This update further solidifies Apple’s competitive edge in edge-side AI by enhancing real-world user experience.

Core Technology Upgrade: Full Support for Gemma4 Visual Path
The standout feature of this update is the native integration of Gemma4’s MTP visual path, the DFlash engine, and ParoQuant quantization technology. Together, these innovations accelerate multi-modal decoding for images and text, drastically lowering the latency barrier for running complex models locally. Consequently, the performance gaps previously associated with local AI have been significantly narrowed through these optimizations.
Usability Leap: omlx Launch Copilot Provides Instant Access to Top Tools
To lower the barrier for both developers and end-users, oMLX introduces the omlx launch copilot. This feature allows users to instantly access leading AI tools like Claude, Codex, and OpenClaw with a single click, enabling seamless interaction between local and cloud services. By enhancing local AI integration, this update makes true "out-of-the-box" usability a reality.
Resource Management Optimization: oQ Smart Proxy Addresses VRAM Constraints
Addressing common deployment challenges within Apple Silicon’s unified memory architecture, the new version features the oQ automatic proxy mechanism. This tool intelligently manages VRAM shortages, ensuring greater stability for large models on consumer-grade hardware. Additionally, the management interface now includes a server restart button, streamlining routine maintenance tasks.
AIbase Commentary: The evolution from MLX to oMLX demonstrates Apple’s rapid progress in edge-side AI, with performance in certain areas now rivaling or exceeding cloud-based solutions. The bandwidth advantages of unified memory, combined with efficient quantization and engine optimizations, highlight the unique benefits of local AI in terms of speed, privacy, and real-time responsiveness. What was once viewed as a limitation in local deployment has now transformed into a remarkable improvement in speed, integration, and usability.
This update sends a clear message: AI is transitioning from the cloud to personal devices. In the near future, more users will experience the freedom and power of running large models directly on their own computers.
Project Repository: https://github.com/jundot/omlx
Related article
48 Hours to TechCrunch Disrupt 2026: Hear From the Builders Shaping What’s Next
With less than 48 hours remaining, TechCrunch Disrupt 2026 kicks off at San Francisco’s Moscone West. Secure your ticket for up to $100 off before door prices rise. Plus, grab a second pass at half price. Register today.From October 13-15, over 10,00
OpenAI Hacking Reports Now With UN AI Safety Council
Yoshua Bengio delivered the opening address at a UN Security Council session on AI safety. Credit: UNReports have emerged of a rogue OpenAI incident involving unauthorized access to sensitive Australian government data, following a UN Security Counci
Anthropic Scientist Quits Over AI Control Fears, Human Self-Inflicted Harm
Artificial intelligence has recently seen a startling departure: Jacob Coxon, a senior researcher at Anthropic, resigned in protest against what he views as an uncontrolled industry frenzy, effectively leaving the AI sector. This move has sparked int
Related Special Topic Recommendations
Comments (0)
0/500
As of May 13, Beijing Time, the Apple Silicon local AI ecosystem has achieved a major milestone. The release of oMLX version 0.3.9.dev2 integrates several advanced optimization technologies, substantially boosting the speed and accessibility of local large language models for image and text tasks. This update further solidifies Apple’s competitive edge in edge-side AI by enhancing real-world user experience.

Core Technology Upgrade: Full Support for Gemma4 Visual Path
The standout feature of this update is the native integration of Gemma4’s MTP visual path, the DFlash engine, and ParoQuant quantization technology. Together, these innovations accelerate multi-modal decoding for images and text, drastically lowering the latency barrier for running complex models locally. Consequently, the performance gaps previously associated with local AI have been significantly narrowed through these optimizations.
Usability Leap: omlx Launch Copilot Provides Instant Access to Top Tools
To lower the barrier for both developers and end-users, oMLX introduces the omlx launch copilot. This feature allows users to instantly access leading AI tools like Claude, Codex, and OpenClaw with a single click, enabling seamless interaction between local and cloud services. By enhancing local AI integration, this update makes true "out-of-the-box" usability a reality.
Resource Management Optimization: oQ Smart Proxy Addresses VRAM Constraints
Addressing common deployment challenges within Apple Silicon’s unified memory architecture, the new version features the oQ automatic proxy mechanism. This tool intelligently manages VRAM shortages, ensuring greater stability for large models on consumer-grade hardware. Additionally, the management interface now includes a server restart button, streamlining routine maintenance tasks.
AIbase Commentary: The evolution from MLX to oMLX demonstrates Apple’s rapid progress in edge-side AI, with performance in certain areas now rivaling or exceeding cloud-based solutions. The bandwidth advantages of unified memory, combined with efficient quantization and engine optimizations, highlight the unique benefits of local AI in terms of speed, privacy, and real-time responsiveness. What was once viewed as a limitation in local deployment has now transformed into a remarkable improvement in speed, integration, and usability.
This update sends a clear message: AI is transitioning from the cloud to personal devices. In the near future, more users will experience the freedom and power of running large models directly on their own computers.
Project Repository: https://github.com/jundot/omlx
48 Hours to TechCrunch Disrupt 2026: Hear From the Builders Shaping What’s Next
With less than 48 hours remaining, TechCrunch Disrupt 2026 kicks off at San Francisco’s Moscone West. Secure your ticket for up to $100 off before door prices rise. Plus, grab a second pass at half price. Register today.From October 13-15, over 10,00
OpenAI Hacking Reports Now With UN AI Safety Council
Yoshua Bengio delivered the opening address at a UN Security Council session on AI safety. Credit: UNReports have emerged of a rogue OpenAI incident involving unauthorized access to sensitive Australian government data, following a UN Security Counci
Anthropic Scientist Quits Over AI Control Fears, Human Self-Inflicted Harm
Artificial intelligence has recently seen a startling departure: Jacob Coxon, a senior researcher at Anthropic, resigned in protest against what he views as an uncontrolled industry frenzy, effectively leaving the AI sector. This move has sparked int











