Meta has released Muse Spark 1.1, an updated version of its Muse Spark model that scores higher than Zhipu AI’s GLM-5.2 on specific coding benchmarks. Artificial Analysis reports the new model achieved a score of 71.3 on the Coding Index, surpassing GLM-5.2 at 68.8 and trailing only GPT-5.6 Luna at 71.4. The Intelligence Index total remains tied at 51 points, but the model gained eight points in just three months, primarily through improvements in coding and agent-based tasks. While top positions remain held by GPT-5.6 Sol and Terra, this performance shift marks a notable gain for Meta’s offering against current competitors.
The update includes practical changes to pricing and operational efficiency that affect deployment costs. Meta estimates the price at $0.26 per task, which is lower than the $0.37 required for GLM-5.2 and significantly cheaper than GPT-5.4 at $0.89. Output token usage also decreased to 94 million from 141 million, while the context window expanded to one million tokens. The hallucination rate dropped from 73 percent to 38 percent, meaning the system now declines to answer more often rather than providing incorrect information. The model is currently available exclusively through Meta’s own API.
- Price per task is $0.26
- Output tokens reduced to 94 million
- Context window increased to one million tokens
