Meta puts Muse Spark 1.3 in its coding tools and drops AI use from performance reviews
Meta is rolling out Muse Spark 1.3 in Muse Code and the Meta Model API, holding the maximum reasoning mode back for further safety testing and keeping an open-weights release on the roadmap. New performance-review guidance drops wording about AI usage and AI Native labels in favour of a line saying results can be supported by AI or other means, and engineers were told that adoption dashboards and token counts would not measure their impact. Employees say their token consumption keeps rising.
Artificial Intelligence··Midday
Muse Spark 1.3 reaches Muse Code while Hatch stays inside the company
Meta says Muse Spark 1.3 is rolling out in Muse Code and the Meta Model API, that the maximum reasoning mode arrives after further safety testing, and that an open-weights release is still on the roadmap. A second agent has not shipped: employees are testing Hatch, an agentic tool that can browse the internet and operate other applications on corporate devices, ahead of an expected public release. Three employees described that testing to WIRED, which reported the internal guidance around it.[1], [2]
Meta's own tables report fewer tool calls and fewer tokens
Meta reports about 20 percent fewer tool calls and about 25 percent fewer tokens on the same work compared with Muse Spark 1.2, and describes stronger resistance to adversarial inputs and prompt injection. Its published comparison tables run against GPT 5.6 Sol (max) and Opus 5 (max) across agent, coding, instruction-following and long-context tasks. All of those figures come from the vendor, and no independent test accompanied the release.[1]
Token counts leave the performance review at Meta
New guidance this week replaces wording about AI usage and AI Native labels with a line saying results can be supported by AI or other means. Three employees told WIRED that evaluations no longer depend on how much they use the company's chatbots and agents, and engineers were separately told that the company would not use AI adoption dashboards or token counts to evaluate impact. A Meta spokesperson said the update restates existing practice and that AI Native labels were never used in evaluations. Employees say their token consumption keeps rising anyway.[2]
