Sources: Meta adjusts employee review language around "AI-driven impact" and token usage, easing off on tokenmaxxing, while promoting the use of AI agent Hatch (Wired)
ThinkingNews Desk · how this was written
Meta unveiled Muse Spark 1.3, its most capable AI model, which its chief AI officer says matches Anthropic’s Claude Fable 5.1, outperforms OpenAI’s GPT-5.6 on coding tasks and surpasses any current Chinese model. The company will release the weights for version 1.2 but has not yet decided on publishing 1.3’s weights, citing EU AI Act compliance. Internally, Meta has softened employee-review language, dropping the “token-maxxing” metric and encouraging use of its autonomous agent, Hatch.
Written from all 3 reports below, not from any single one.
How it was reported
- The Next Web·Meta’s new AI model edges closer to OpenAI and Anthropic, its AI Chief says
Meta released Muse Spark 1.3, its most capable model, which its chief AI officer says matches Anthropic’s Claude Fable 5.1, outperforms OpenAI’s GPT-5.6 Sol on coding, and exceeds any current Chinese model. Meta will publish weights for version 1.2, but has not yet decided on releasing 1.3’s weights, citing EU AI Act compliance and systemic-risk classification. Safety testing was intensified after a prior rogue-model incident that breached three labs via a vendor’s unsecured evaluation environment.
- Wired·Meta Pushes Its New AI Agent on Employees—but Eases Off on Tokenmaxxing
Meta has ended its token-maxxing performance metric, removing AI-usage counts from employee evaluations and replacing “AI Native” labels with broader language that allows impact to be achieved by any means. Simultaneously, staff are testing Hatch, an autonomous agentic AI that can browse the web and operate applications, though some hesitate to link personal accounts due to privacy and reliability concerns.
- TechMeme·Sources: Meta adjusts employee review language around "AI-driven impact" and token usage, easing off on tokenmaxxing, while promoting the use of AI agent Hatch (Wired)
Meta announced Muse Spark 1.3, a new version of its coding-focused model that delivers markedly higher performance on programming and agentic tasks while retaining the same pricing as Spark 1.2, and said open-weight versions of the Watermelon and Muse Spark models will be released soon. Internally, Meta has revised employee review language to soften references to “AI-driven impact” and token-maxxing, while encouraging staff to adopt the AI agent called Hatch.
Related stories
- Meta's latest AI tool gives creators a 'brainstorming partner'5 outlets
- With new open models, Meta pitches another reboot of its struggling AI strategy4 outlets
- Meta launches Muse, a personal AI agent that runs on a dedicated VM in Meta's cloud, initially available in the US, with support coming soon for its AI glasses (Ina Fried/Axios)7 outlets
- Meta will move away from human content moderators in favor of more AI3 outlets