Meta released Avocado, named Muse Spark

Meta released Avocado, they call it Muse Spark.

Meta released Avocado under the name Muse Spark; the post says TBD lab rebuilt the pretraining stack in 9 months and reached capability similar to Llama 4 Maverick with over 10x less compute. The post also says it is not open source; the post does not disclose model size, benchmarks, parameter count, or release timing. The real signal is infrastructure efficiency, not the rename.

Why it matters: HKR-H/K/R all pass: the real hook is a near-Maverick claim at under 1/10 training compute after a 9-month stack rebuild, and infra efficiency hits a core cost nerve. Held at 74 because the post does not disclose params, benchmarks, or release timing, and this is still a single-sr