As of Oct 5, MaruthLabs launched Madhuram-v1, a 151M-parameter on-device model.
Timeline · newest first
Oct 5
Official
MaruthLabs launches Madhuram-v1, a 151M-parameter on-device model
MaruthLabs launched Madhuram-v1 on October 5: a 151M-parameter model pretrained on 1 trillion tokens with 16K context, aimed at on-device use. The company says it beats LFM2.5-230M, Gemma 3 270M and other peers on 9 of 17 pretraining benchmarks — comparisons from its own suite, without third-party reproduction.