Superpower Daily
micro1 Bids $12.5 Million for Spirit Records, Challenging Googl...Artificial Analysis Clarifies Astra’s 67-Point Coding Score Is...Anthropic’s Claude Built a 13-Million-Line Fermat Proof, Leavin...Microsoft Opens MAI-Image-2.6 in Foundry Preview and Adds a Low...Databricks Says Proteus Generated Qwen Kernels Up to 5.2× Faste...Lutnick Ties G20 AI Infrastructure Pitch to U.S. Tariff Reliefmicro1 Bids $12.5 Million for Spirit Records, Challenging Googl...Artificial Analysis Clarifies Astra’s 67-Point Coding Score Is...Anthropic’s Claude Built a 13-Million-Line Fermat Proof, Leavin...Microsoft Opens MAI-Image-2.6 in Foundry Preview and Adds a Low...Databricks Says Proteus Generated Qwen Kernels Up to 5.2× Faste...Lutnick Ties G20 AI Infrastructure Pitch to U.S. Tariff Relief
Modelspublished

Meta Adds Max Reasoning to Muse Spark 1.3 After a Two-Day Safety Gate

The setting behind many of Meta’s strongest reported evaluations is now usable in Muse Code and its API, though the company’s efficiency and performance claims remain largely its own measurements.

By 2 min read
Meta Adds Max Reasoning to Muse Spark 1.3 After a Two-Day Safety Gate
Meta Adds Max Reasoning to Muse Spark 1.3 After a Two-Day Safety Gate

Listen to this story

The audio brief

About 1:26
0:001:26
Read transcript
Meta has released the Max reasoning setting for Muse Spark 1.3, just two days after holding it back for additional safety testing. Max is now available in Muse Code and through the Meta Model API, so developers can use the configuration behind many of Meta’s strongest reported evaluations—not just test it in isolated prompts. Muse Spark 1.3 launched on September second with lower reasoning modes available first. Meta positions the model for sustained coding and agent work: managing several workflows in one thread, gathering context with tools, repairing gaps in a plan, and retaining information for a final deliverable. It is also designed to ask clarifying questions when a request is ambiguous, seek help instead of claiming success when it gets stuck, and confirm before taking consequential actions. The important caveat is that Max does not yet come with its own public pricing or latency figures. Meta says the broader Muse Spark 1.3 release used about twenty percent fewer tool calls and twenty-five percent fewer tokens than Muse Spark 1.2 in internal comparisons. Those numbers are Meta’s measurements, and they do not establish how Max performs on cost or speed specifically. The model remains proprietary. Meta says an open-weights release is planned, but has not named a version or date. For now, the real test is whether developers use Max for long-running work in Muse Code and the API, rather than only for difficult one-off tasks.

Story brief

3 key points

Meta has completed the staged rollout of Muse Spark 1.3’s Max reasoning mode, adding it to Muse Code and the Meta Model API after two days of extra safety testing. Max is the configuration behind many of the model’s strongest reported evaluations, but Meta has not provided separate pricing or latency results for it. The proprietary model targets sustained coding and agent workflows, while a planned open-weights...

  1. 01

    Muse Spark 1.3 launched September 2 with lower reasoning modes; Max was held back pending additional safety testing.

  2. 02

    Meta reports roughly 20% fewer tool calls and 25% fewer tokens than Muse Spark 1.2 in internal comparisons.

  3. 03

    Max is available through Muse Code and the Meta Model API, expanding access beyond isolated prompt experiments.

Meta said maximum reasoning for Muse Spark 1.3 would follow additional safety testing. Two days later, the company released the Max setting in Muse Code and the Meta Model API, giving developers access to the configuration used for many of its strongest reported results.

The change completes a staged rollout: Meta introduced Muse Spark 1.3 on September 2 with other reasoning modes available first, while Max remained behind a safety gate. The setting now brings the highest reasoning level to the coding tool and API used by developers.

The agent-work pitch

Meta positions the model for long-running coding and agent work rather than isolated prompts. It says Muse Spark 1.3 can manage several workflows in one thread, use tools to assemble context from conflicting sources, repair gaps in a plan, and retain information for a final deliverable.

Meta says the model is designed to

  • Ask clarifying questions when a prompt is ambiguous.
  • Seek help from the user when it gets stuck rather than claim an outcome it cannot produce.
  • Confirm before consequential actions and adapt to requests for progress updates or silent background work.

What Max does—and does not—show

Meta used Max for many of Muse Spark 1.3’s strongest reported evaluation results. In internal comparisons, the company said Muse Spark 1.3 used about 20% fewer tool calls and 25% fewer tokens than Muse Spark 1.2—figures about the model release, not a complete pricing or latency analysis for Max.

A proprietary developer route

Muse Spark 1.3 remains proprietary. Meta says it plans an open-weights Muse Spark release but has not identified a version or date; the immediate test is whether developers use Max for sustained work in Meta’s coding tool and API rather than only for difficult one-off tasks.

Sources

  1. research.meta.aiIntroducing Muse Spark 1.3

Related coverage

Continue reading