RL agent reportedly makes $85,771 trading a real Texas battery, 4x the human operator's $20,647
xeophon · x · 2026-11-05
ScaleAI Labs built a high-fidelity RL environment simulating a trading desk on a real Texas battery grid, wired into Prime Intellect's prime-rl framework via the AgentEnv framework.
In the showcased run, an agent dubbed GPT-6.1 Sol earned $85,771 trading a real Texas battery, while the battery's actual human operator made only $20,647 over the same period — roughly 4x less.
The AgentEnv docs also lay out the full integration: prime-rl trains through verifiers' NeMo Gym taskset, which seeds sessions from dataset rows, calls tools, and verifies rollouts for rewards. Setup requires Linux with at least two NVIDIA GPUs (one to train, one to serve via vLLM), with complete install and deployment instructions provided.
More from coding & agent
- Weaviate Podcast: Forgetting Is the Key to Scaling Agent Discovery, With Berkeley's Shangyin Tan — CShorten30 · 2026-11-05
- Dots Team Recruits to Build Out Persistent Agents — sherwinwu · 2026-11-05
- Managing agent workstreams as infrastructure-as-code in production — JoshPurtell · 2026-11-05
- AI Coding Tool Conductor's Team Joins Meta to Build the Next Interface — ycombinator · 2026-11-05
- 'You own every line you submit': the coding policy that survives AI — esrtweet · 2026-11-05
- Notion AI agents gain Google Drive access: search files, edit Drive, trigger workflows — technextpreneur · 2026-11-05