Meta's Muse Spark 1.1 Outperforms GLM-5.2 in Coding Benchmarks and Lowers Developer Costs
Following its global rollout, Meta's newly launched Muse Spark 1.1 has emerged as a highly competitive, low-cost option in the developer market, directly challenging Z.ai's open-weight Z.ai Unveils GLM-5.2 and ZCode Agentic Environment Following Massive $4 Billion Capital Raise and proprietary US models. According to the latest evaluation by Artificial Analysis, Muse Spark 1.1 scored 51 on the main Intelligence Index, tying with GLM-5.2, GPT-5.4, and GPT-5.6 Luna. This represents an eight-point gain in just three months, driven primarily by improvements in coding and agentic reasoning workloads.
Superior Coding and Cost-Performance Efficiency
On the Artificial Analysis Coding Index, Muse Spark 1.1 scored 71.3, edging past GLM-5.2's score of 68.8 and landing just a hair behind OpenAI's GPT-5.6 Luna (71.4). The top positions on the Index remain occupied by OpenAI's GPT-5.6 Sol (77.4) and Terra (76.7), followed by Anthropic's Claude Fable 5 (76.5).
The commercial appeal of Muse Spark 1.1 lies in its superior cost-efficiency for developer pipelines:
- Muse Spark 1.1: Estimated $0.26 per task, utilizing only 94 million output tokens.
- GLM-5.2: Estimated $0.37 per task, utilizing 141 million output tokens.
- GPT-5.4: Estimated $0.89 per task.
In addition to the lower cost per task, Meta has successfully reduced the model's hallucination rate from 73% to 38%, largely by training the model to decline to answer rather than providing incorrect information. Meta has also quadrupled the model's context window to 1 million tokens. At launch, Muse Spark 1.1 is available exclusively through Meta's proprietary API, serving as a key lever to attract developers directly to Meta's infrastructure.