Cloudflare used its own variant of Reinforcement Learning for Calibrated Decisions to train Clef, a method that TypeSafe AI previously used for Jev

Overview Debates Related claims Topics Sources

DeepSeek trained DeepSeek-V4.1-Flash-1017 using reinforcement learning with self-feedback to improve performance on end-to-end coding, game development, and 3D modeling tasks

DeepSeek's reinforcement learning with self-feedback method uses internal confidence scores and self-generated critiques as reward signals, eliminating reliance on human evaluators or external reward models

Reflection AI released post-training recipes, code templates, and safety measurement tools alongside Beam's weights on October 16, 2026, to enable developers to fine-tune the Beam model

Reflection AI trained Beam using intensive reinforcement learning that penalized the model whenever it generated unnecessary tokens to complete a task

Reflection AI developed automated reliability optimization algorithms that detect data quality errors and revert training runs to earlier stages instead of restarting from scratch

External partner Cinder evaluated both FLUX 3 Image Klein checkpoints for safety prior to the model's release following dataset filtering and post-training to limit harmful outputs

Post-training Nemotron 3.5 Super VL on specialized domain data improved model accuracy by 9 to 44 percent across document, cybersecurity, finance, science, and medical tasks

Organizations including Scale AI, EY, CrowdStrike, Ramp, and AI21 are post-training Nemotron 3.5 Super VL for specialized enterprise workflows such as audit, credit analysis, and cybersecurity

QwenGyre achieved end-to-end training speedups of 1.21x to 1.85x over colocated setups and 1.38x to 1.78x over asynchronous setups across benchmark evaluations.

QwenGyre's Elastic Scheduler dynamically transfers GPUs between rollout generation and model training without interrupting active agent executions.

Applied to the Qwen 3.8 2.4T model on NL2RepoBench, QwenGyre increased accuracy from 52.5% to 58.5% in 48 training steps, completing the steps in 75.42 hours compared to 134.55 hours for asynchronous training.

Alibaba Qwen, in collaboration with the University of Science and Technology of China and Tsinghua University, introduced QwenGyre, an elastic reinforcement learning framework for training extreme-long-horizon agents.

QwenGyre trains coding agents through unmodified black-box harnesses like Claude Code 2.1.220, using Megatron for model training and SGLang for rollout serving across multi-node GPU clusters.

QwenGyre's Trajectory Processor caps retained trajectories at 5 per execution, cutting forward-backward training time by 25.2% while matching uncapped score performance

In the QwenGyre framework, satellite GPU cells migrate key-value cache blocks over RDMA and pull authoritative weights from a core cell via Mooncake's Transfer Engine when joining training batches.

QwenGyre's Elastic Scheduler recorded average role switching times of 8.52 seconds from rollout to training and 3.46 seconds from training to rollout.

The censorship in Qwen is baked into the model's core training protocols through supervised fine-tuning and reinforcement learning from human feedback, rather than an external filter.

Cloudflare is introducing a reinforcement learning fine-tuning platform and service to let customers customize Cloudflare's Clef for their own tasks

Reflection AI used 10.5K NVIDIA GB300 GPUs for four weeks to run high-compute reinforcement learning for its open-weight model Beam, generating over 100 million rollouts

Aleph Alpha's development teams in Germany controlled the entire pipeline for the Kolibri model, including data, architecture, training infrastructure, post-training, and evaluation