Open-source Gradients is progressing nicely... But we're impatient and want to get the sign-off faster. This week we'll be updating the tourn structure to allow more competitors and faster tournament iterations, plus a revamp to the incentive mechanism to continue to

Right now, only a subset of the mining pool can take part in Gradients open-source training competitions. We decide who based on alpha stake - a proxy for motivation level. From next week we'll double the number of miners that can join. Then soon after, a revamp is coming.

Gradients Instruct 8B is the most widely used model ever trained on Bittensor. It has been used over 3.3M times in the past 7 days through Chutes Try it here: chutes.ai/app/chute/280c… Want to train your own model? Get started in seconds on gradients.io

Everyone thinks AI is controlled by a few closed labs like OpenAI or DeepMind Wrong. Gradients ( Subnet 56) is a global arena where thousands of independent minds battle in real-time tournaments, building models that evolve faster than billion dollar R&D And the craziest

The results on SN56 speak for themselves: 🏆 Beat Google, Databricks, Huggingface, Civit and TogetherAI in training abilities 📄 arxiv.org/abs/2506.07940 🏆 Outperformed Qwen and Meta with Gradients + our in-house data pipeline across countless 8B benchmarks 📊 3.8

Subnets competing in Subnets. 🤖 Gradients miners automatically producing competing models for Affine - launching end of week 🔁 Synthetic data generation based on Affine reward functions for GRPO model training 📊 10% of all GRPO tasks dedicated to Affine-compatible

Really excited to share that open-source miners on Gradients have BEATEN closed-source Gradients miners for the first time. This isn't across the board yet, but it's an incredible step forward. Once we achieve wins across the board, we won't just have the best open-source

The week ahead: 🚀 Multi-dataset scheduling - breaking training into stages to handle millions of datapoints - all with still just a few clicks 📊 Internal benchmark reveal - open-source progress overtime charts dropping 🎯 Gradients 33B - preparing to make waves 📽️ Video

Today our paper was accepted at Distributed AI (DAI) conference 2025 . arxiv.org/pdf/2506.07940 Research validation: Gradients is the best AutoML training platform on the planet.

An amazing breakthrough today at Gradients The image open-source tournament winner just hit performance parity with the best closed-source training platform on earth (also Gradients 😉) 🏆 The latest winner completely dominated ⚡ Insane flux fixes and optimisations +

New to Gradients? This walkthrough shows the exact sign-up path. Fast, straightforward, and designed so anyone can train.

Train on demand. Monitor in real time. This walkthrough takes you from zero to a tuned model you can actually use.

We just published the full story of how open-source tournament miners reached parity with closed-source. The economics, the timeline, the future of Gradients. Read it here: medium.com/@samoline/open…

Big week ahead for Gradients with a major codebase update. → 6.0 - upgraded incentive mechanism + privacy-focused hosted service for customers → Gradients V2 model benchmarking auto-integrated into our in-house pipeline → TAO payment integration for customers goes live