Yandex ML Challenge opens May 21 with 1M ruble prize across four tracks

Yandex ML Challenge registration closes this week ahead of a May 21 start. The competition runs an 11-day online round across four tracks: large language models and foundation models, computer vision, reinforcement learning, and neural network optimization. Each participant gets 40 submissions per task with no hard time limit during the online phase, letting competitors iterate on approaches without the pressure of a timed sprint.

The top 100 scorers advance to an in-person final at Young Con in Moscow. First place takes 1 million rubles, and the top 15 finishers receive hardware from Yandex. The format skews toward practitioners who want to test production-relevant techniques rather than pure speed-coding—foundation model tuning and RL optimization both demand longer iteration cycles than the typical four-hour hackathon allows.

The competition draws from Yandex's internal ML stack, so tasks likely reflect real deployment constraints: inference budgets, multi-modal pipelines, model compression for edge cases. That makes it a useful benchmark for anyone working outside pure research environments, where compute is finite and latency matters. Watch for leaderboard movement in the first 48 hours—early submissions often reveal which tracks have the tightest baselines and where novel approaches can still move the needle.