DeepSeek V4 Models Become Default for OpenClaw Open-Source Agent Framework
V4 Flash delivers near-Pro performance at a fraction of the cost
The global open-source agent framework OpenClaw released its 2026.4.24 version on Sunday, integrating both DeepSeek V4 models into its core ecosystem. V4 Flash, featuring 284 billion total parameters with only 13 billion activated per inference, is now the default model. It delivers Max mode inference performance nearly matching the 1.6 trillion parameter V4 Pro, making high-end agent capabilities more accessible without the full resource overhead.
Beyond model upgrades, the update introduces real-time voice integration with Google Meet, allowing agents to participate in live conversations. Enhanced browser automation recovery improves reliability during complex web tasks, while resolved issues in multi-step tool execution streamline agent workflows. V4 Pro, recognized as the world’s largest open-source model, is also available in the model library for users needing maximum capability.
- DeepSeek V4 Flash has 284B total parameters but only 13B activated, achieving near-Pro performance
- V4 Pro (1.6T parameters) is the world’s largest open-source model, now in OpenClaw’s library
- Update adds real-time Google Meet voice, better browser recovery, and multi-step tool fixes
Why It Matters
Open-source agents can now rival proprietary models, lowering costs and expanding developer access to advanced AI.