What happened
The GitHub release page for google-deepmind/mujoco lists version 4.0.1, released on November 11, 2026. The update includes a compiler API addition, a Python tooling change and several bug fixes. The release notes do not describe a new major version or a change to MuJoCo’s overall positioning; they frame this as a point release addressing compatibility and correctness issues.
For Python users, the release enables the MuJoCo Studio Web Viewer on Windows. The notes identify two ways to launch it: mujoco-studio --gfx=web and mujoco.studio.launch_web. The Windows support is delivered by a precompiled browser client in a platform-independent mujoco-studio-wasm wheel, which is installed automatically by mujoco-studio. That packaging detail means users do not need to install the browser client separately, according to the release notes.
The most consequential correction concerns sparse Cholesky factorization. The release notes say the bug was introduced in 4.0.0: rows with different descendants could be merged into one supernode. When the constraint Jacobian is sparse, this could give the Newton solver an incorrect Hessian factor, resulting in slow convergence, wrong accelerations or divergence. The release associates the fix with issue #3767. These are potential numerical and simulation-quality problems, rather than merely cosmetic output differences.
The release also adds mjs_getInStringVec, an API for reading entries from an mjStringVec by index. It changes plugin-library loading so mj_loadAllPluginLibraries skips a dynamic library that fails to load and emits a warning rather than calling mju_error. The notes say that, in the 4.0.0 macOS and Linux archives, this issue affected the USD decoder plugin, which can only be loaded from a Python process. Finally, 4.0.1 fixes single-shot multicontact issues involving cylinder edge collisions. The release page does not state a price or separate access restriction for the update.
What the archive adds
New Runtime covered MuJoCo 4.0’s Studio viewer, Filament rendering, deformable simulation and speculative contacts on November 10, 2026. Version 4.0.1 is a new fact on that story: it extends the Studio Web Viewer to Windows and corrects a numerical bug introduced by the release covered the day before.
Trend signal
This release is a weak fit for agent economics moves to completed work: more reliable simulation can matter to systems judged by task outcomes, but the release does not report agent use, completed-task rates or costs. This does not yet fit a pattern New Runtime tracks.
Why now
MuJoCo 4.0.1 followed the 4.0.0 release by one day, according to the release dates and New Runtime’s earlier coverage. The notes explicitly identify the solver defect as introduced in 4.0.0, making this a prompt correction to a regression in the new major version. The page does not explain the schedule beyond those release facts.
Connections to other events
The 4.0.1 fixes meet the broader 4.0 launch covered in New Runtime’s earlier coverage. The connection is direct: the release notes say the sparse factorization bug was introduced in 4.0.0, while the Web Viewer had been among the new Studio capabilities discussed in that earlier coverage. This update extends that workflow to Windows and repairs a regression in the new release.
Why it matters
For teams building robotics, control or embodied-AI workflows on MuJoCo, a solver defect can affect the validity of the simulation itself. Slow convergence may make experiments harder to run; wrong accelerations or divergence can undermine results or interrupt a pipeline. Teams using sparse constraint Jacobians on 4.0.0 should treat the fix as more than routine housekeeping and check whether their scenarios exercise the affected path.
Windows support for the Web Viewer also removes a platform limitation for users who want a browser-based Studio workflow. The automatically installed, platform-independent wheel reduces a packaging step, though the release does not claim broader performance or workflow improvements. The plugin-loading change is relevant to startup resilience: one unloadable library now produces a warning rather than aborting simulate at startup. Together, these changes make the point release useful to existing users, but they do not establish that every MuJoCo deployment is affected by the solver bug or needs an immediate migration.
Limits and uncertainty
The release notes describe the failure modes but do not quantify how often the sparse factorization bug occurred, identify which models or workloads were affected, or report numerical comparisons before and after the fix. They do not establish whether users’ published results were affected. The page also provides no benchmark for Windows Web Viewer performance or compatibility information beyond the stated enablement. New Runtime has not independently tested the fixes; teams should validate their own simulations and startup paths after upgrading.
Skill to build now
If your team uses MuJoCo 4.0.0 with sparse constraint Jacobians, upgrade a representative environment to 4.0.1 and rerun simulations that exercise those constraints. Compare convergence, accelerations and completion against a trusted baseline, and record the MuJoCo version alongside experiment results. If you use Studio on Windows, check the documented web launch commands and confirm the automatically installed mujoco-studio-wasm wheel is present.
Reader takeaway
MuJoCo 4.0.1 is a focused follow-up to 4.0.0: it brings the Studio Web Viewer to Windows and fixes a sparse-solver regression with potentially serious numerical consequences. The practical priority is highest for teams using sparse constraints on 4.0.0; validate those workloads rather than assuming every simulation is affected.