SpaceXAI releases Grok Build v1.0.41 update

Looking at the v1.0.41 change list as a whole, the main thread is not hard to identify: it deals almost entirely with problems that only surface when things 'run long, run a lot, and run scattered.' Sub-agents inheriting model settings, persistent configuration, per-model request size limits, long-reasoning reminders, MCP server configuration handling, and crashed sessions showing interrupted turns—these items each look trivial on their own, but they point to the same usage scenario: multiple sub-agents collaborating over long periods.

The publisher himself gave the same assessment, saying that 'a lot of the work here is clearly focused on making long-running multi-agent workflows more reliable.' It should be noted that this is the release post author's commentary, a judgment from the source side rather than an independently verifiable conclusion; however, the change list itself aligns with this direction, so it can be regarded as a reasonable summary of the list.

The only end-user-facing new capability in the list is the sports_Search tool. According to the release post, it is used to obtain NFL live scores, rankings, schedules, player data, and team records, with data coming directly from 𝕏.

There are two points here that need to be viewed separately. First, the tool's existence and purpose are explicitly listed in the release post; second, the 'live' attribute and data quality have not been given any independent verification or metrics by the source, so they can only be understood as the source's claim and should not be treated as a confirmed level of capability. In addition, the list does not state whether the tool covers leagues beyond the NFL, nor does it give any usable scope or regional restrictions.

Sub-agents can now inherit model settings and support persistent configuration. One comment in the replies directly pointed out the practical value of this change: 'Inheriting settings is nice, nobody wants to configure the same thing five times.' This comment comes from a user reply on X and is a bystander's evaluation, but it matches the problem the change itself is meant to solve.

Paired with this is the per-model request size limit. The release post says its purpose is to better control vendor limits and inline images. Putting these two items together, it can be inferred (this is an inference) that when dealing with concurrent multiple sub-agents, configuration consistency and request size are two main friction points for the team; however, the source does not give any specific limit values, nor does it state default values or adjustable ranges.

The list also includes two changes leaning toward 'process control': first, a new long-reasoning reminder, used to keep the model on track during long reasoning processes; second, improved MCP server configuration handling.

Neither of these comes with implementation details or performance data. What can be confirmed is that they are listed as important changes in this release; as for the form in which the reminder mechanism is triggered and what exactly changed in MCP configuration handling, the source does not explain. Therefore, this can only remain at the level of 'listed in the change list.'

Among the reliability-related items, the most specific one is: crashed sessions now clearly display the interrupted turn instead of silently losing it. This change is worth calling out separately because it turns a previously invisible type of failure into a visible state.

The reaction in the comments to this change was also the most direct, with a user asking how long 'silently losing sessions' had existed as a shipped feature. This question itself does not constitute confirmation of historical versions; it only shows that the issue resonated on the user side. The source does not explain the specific behavior details of previous versions, nor does it provide data such as crash rates or recovery success rates.

The list finally includes a batch of fixes involving sub-agent generation, queued prompts, interjections, edit blocks, and effort level selection. These items also lack details; it can only be confirmed that they fall within the scope of fixes in this release.

It needs to be made clear that all of the above content comes from a single release post on X and its comment section. There is no first-party release note, version log, or independent testing for cross-verification. Therefore, the version number v1.0.41 and the content of each change are statements from the source side; the publisher's summary about 'focusing on long-running multi-agent reliability' is a judgment from the source side; and regarding the actual effects of these changes, there is currently no verifiable basis.