Meta continues to aggressively reshape the developer model market, turning intellectual labor into a low-cost commodity. According to recent data from Artificial Analysis, the new Muse Spark 1.1 gained eight points this quarter, hitting 71.3 on the Coding Index. This leap was enough to overtake China’s GLM 5.2 (68.8) and put it within striking distance of GPT-5.6 Luna’s 71.4. While high-end models like GPT-5.6 Sol and Terra maintain their lead at 77.4 and 76.7 respectively, Meta has achieved parity in "raw intelligence": Muse Spark 1.1, GLM 5.2, and Luna now share a matching score of 51 on the Intelligence Index.
The real drama, however, is unfolding in the accounting department rather than the benchmarks. Meta is undercutting the competition with surgical precision: the cost to complete a task with Muse Spark is just $0.26. In comparison, China’s GLM-5.2 charges $0.37 for the same workload, while OpenAI has set what is effectively a prohibitive rate of $0.89 for GPT-5.4. The efficiency of Meta’s architecture is equally impressive—the model requires only 94 million output tokens where GLM "burns" through 141 million. Reliability has also improved significantly, with hallucination rates dropping from 73% to 38%, largely because the model has learned to admit when it lacks knowledge instead of hallucinating with confidence.
Technological Breakthroughs and Market Challenges
The context window has expanded to one million tokens, quadrupling the capacity of the previous version. Hallucination rates have been nearly halved, falling from 73% to 38%. Architectural optimization allows for 33% fewer tokens consumed per task compared to competitors.
Mark Zuckerberg’s strategy is clear: if you cannot win the race for absolute superintelligence right now, make your competitors' intelligence economically unviable.
Yet, questions remain beyond these sterile tests. Scoring high on indices is one thing; performing within real-world AI agent workflows is another, where businesses still gravitate toward Claude Fable 5. Meta is betting on the democratization of high-level reasoning, forcing competitors to justify their premium pricing with more than just brand recognition. While OpenAI struggles to recoup its massive electricity bills, Meta is turning AI coding agents into a mass-market commodity with price points that leave little room for high margins among rival model developers.