Meta's Muse Spark 1.1 scored 51 on the Artificial Analysis Intelligence Index, up eight points in three months. It edges past GLM-5.2 in coding with a score of 71.3 at a lower cost of $0.26 per task, and its hallucination rate dropped from 73 to 38 percent.
SynthePulse Insight · AI deep reading
Version 2 · 1 source
Meta's Muse Spark 1.1 edges past GLM-5.2 on the Artificial Analysis coding index with a score of 71.3 versus 68.8, while costing less per task and cutting its hallucination rate sharply. This update shows that AI model competition is shifting from pure scale toward iteration speed and specialized optimization.
According to Artificial Analysis data, Meta's Muse Spark 1.1 scores 71.3 on the coding index, slightly ahead of GLM-5.2's 68.8 but behind GPT-5.6 Luna's 71.4. The top spots are held by GPT-5.6 Sol (77.4), Terra (76.7) and Claude Fable 5 (76.5). This gap indicates that Muse Spark 1.1 has entered the top tier in coding specifically, but does not hold an absolute lead.
On cost, Muse Spark 1.1 spends an estimated $0.26 per task, below GLM-5.2's $0.37 and GPT-5.4's $0.89. At the same time, it uses only 94 million output tokens, whereas GLM-5.2 uses 141 million. This cost efficiency could make it more attractive in price-sensitive application scenarios.
Muse Spark 1.1's hallucination rate dropped from 73% to 38%, meaning the model more frequently refuses to answer rather than providing incorrect information. This improvement directly raises the model's reliability in scenarios such as knowledge work.
In addition, Meta expanded the context window to 1 million tokens, four times the previous size. This allows the model to handle longer documents and more complex conversations, but the actual effect still needs verification.
On the Artificial Analysis intelligence index, Muse Spark 1.1 scores 51, tied with GLM-5.2, GPT-5.4 and GPT-5.6 Luna. Over the past three months, the model's intelligence index rose by 8 points, mainly from coding and agent-based knowledge work.
This pace of progress suggests that Meta is improving its competitiveness through rapid iteration rather than simply scaling up. However, benchmark scores do not always reflect real-world performance, and the gap in actual applications may be larger.
Currently, Muse Spark 1.1 is offered only through Meta's own API, which limits its accessibility. The evaluation data comes from the third party Artificial Analysis and may be constrained by specific benchmarks, and long-term availability and pricing may change.
In addition, models such as GLM-5.2 may be underrated in Chinese-language scenarios, and the connection between benchmark results and real-world applications has not been fully verified. Therefore, although Muse Spark 1.1 performs well in coding and cost, its overall advantage still needs further observation.
This article is based on THE DECODER's reporting on Artificial Analysis data; all performance metrics are third-party evaluation results, and actual performance may differ.
Muse Spark 1.1 has made significant progress in coding performance and cost efficiency, but its narrow lead and the limitations of third-party evaluation mean its real-world advantage still needs verification. AI model competition is shifting toward iteration speed and specialized optimization.
Primary report
Primary source