Google Gemini’s Delayed Flagship Model Sparks Mockery as AI Throne Battle Heats Up

When Meta’s chief AI officer Alexandr Wang posted "gemini who?" alongside a model‑performance leaderboard, it ignited a public jab at Google’s Gemini series, turning a technical comparison into a broader power shift in the AI arena.

The Artificial Analysis Intelligence Index ranking, shared by X user Tae Kim, placed Meta’s Muse Spark 1.1 ahead of Google’s Gemini 3.6 Flash in key dimensions such as reasoning, code generation, and agent task execution, with Muse Spark scoring significantly higher while Gemini 3.6 Flash received a modest 50 points.

Google’s flagship Gemini 3.5 Pro has suffered repeated postponements, weakening its market stance. In response, Meta highlighted its strategic emphasis on agent‑level autonomy, code‑level proficiency, and real‑world task execution, arguing that raw benchmark scores are no longer the sole success metric.

Critics like François Chollet have accused Muse Spark of over‑optimizing for public benchmarks, suggesting a gap in generalisation. Meta countered that while Muse Spark underperforms on tests such as ARC AGI 2, its focus remains on advancing agent, coding, and practical task capabilities.

Google’s Gemini 3.6 Flash does show progress: on the DeepSWE code test its score rose from 37 % to 49 %, and on the OSWorld‑Verified computer‑operation test it improved from 78.4 % to 83 %, indicating both firms are moving in the same direction despite differing public narratives.

Analysts weigh in: Raymond James’ Josh Beck notes that Gemini 3.5 Pro’s delays have turned Google from a market leader into a chaser, while Moor Insights’ Anshel Sag argues that Google’s massive cloud infrastructure and data assets constitute a durable moat, even as it pre‑announces Gemini 4.

The episode underscores a harsh reality in the AI race: the throne is never secure. Months ago Google reclaimed advantage with Gemini 3, yet today it faces coordinated attacks from Meta, OpenAI, and Anthropic, all prioritising agent‑first strategies, proving that no AI leader can remain unchallenged for long.