L4 launched without a thinking model, making it an inferior choice for coding, one of the main LLM use cases. Even in benchmarks it wasn't even competitive at coding with 3-month-old Deepseek R1.
Agreed, coding is not a strong point of L4. However the "hive mind" in some places thinks L4 is "a failure" and "useless". In reality it is an "ok" model, and most 3rd party benchmarks done after the inference lib updates were in line with whatever meta announced.
