exist DeepSeek V4Two days after its release, OpenClaw received another update, completing the adaptation and access to this series of models.Early this morning, OpenClaw released version ‘2026.4.24’.


As can be seen from the detailed updated interface, this time OpenClaw pays more attention to productivity integration and model architecture: it is no longer just a simple conversation, but deeply embedded in your workflow.

Highlights of the update include the following:

One is,Google Meet deep integration.

OpenClaw is now available as a plug-in directly into Google Meet. It supports personal account authorization, real-time session handling, and even handles open meeting tabs. The most powerful thing is that it supports output and attendance export, and has an automatic recovery mechanism, completely turning AI into your "meeting secretary".

Second,DeepSeek V4 fully accessible.

DeepSeek V4 Flash is the default preferred model, and the Pro version is launched simultaneously. More importantly, the official has fixed the thinking logic of DeepSeek in consecutive rounds of tool calls, making the response smarter.


The third is,Voice calls are open across the board.

Whether it’s a normal conversation or a voice call (including Meet), real-time voice loops are now supported. This means that AI is no longer just a repeater, but can call all tool plug-ins in the OpenClaw backend to provide you with in-depth, well-documented answers.

Fourth,Browser automation evolution.

Coordinate clicks are now supported, and the upper limit of operation time (Action budgets) has been extended. At the same time, the browser's tab reuse and crash recovery have become more stable, and it also supports the independent setting of 'headless mode' for different configuration files.

Fifth,Detail polishing and restoration.

Special fixes have been made for Telegram, Slack, MCP protocol, session management, and TTS (speech synthesis) to further improve system stability.

Sixth,‘Downsizing’ the underlying architecture.

The startup speed has been optimized, and the model directory and plug-in loading have been changed to 'lazy loading', so that the memory will no longer be occupied as soon as it is started. Dependency repair of the packaged installation version has also become simpler.

In addition, OpenClaw this time alsoRefactored plugin SDK.

In order to ensure consistency under different environments (Pi and Codex), the old compatibility path has been officially deleted. All tool result overrides must now go through api.registerAgentToolResultMiddleware and be explicitly stated in the contract. To put it simply: the code is more standardized, but it also requires developers to adapt to the new middleware logic.

At the same time, DeepSeek officially updated the API documentation and announced that DeepSeek-V4-Pro is offering a limited-time 25% discount, with an expiration date of May 5.It also mentioned that OpenClaw needs to be upgraded to the latest version in order to use the latest series of models.

After the price adjustment, the input price of the model (cache hit) is 1M tokens/0.25 yuan, the input price (cache miss) is 3 yuan, and the output price is 6 yuan.


Everyone is very curious about the performance of the DeepSeek V4 series.


Hot comments from netizens: Stop playing, withdraw

OpenClaw's update cadence has been a mixed bag.

As the netizen below said, ‘This update brings huge changes, including real-time voice, Google Meet agents, faster startup speed, and better local models. ’


However, more netizens have ‘broken their guard’.

Many people reported that after the update, the entire system crashed and the core 'gateway' could not be started at all.



Someone else said, 'Another update full of bugs. Brothers, I won’t play with you anymore. I’m very busy and have to mess with my Hermes intelligent agent that can run through. ’