Summary
According to a post on X by @samsja19, Prime Intellect has released prime-rl 0.11.0, an open-source reinforcement-learning trainer. The update adds a new pretraining entry point, faster training with model-specific kernels, two-way KV-cache transfer and lower CPU memory use. The post also says coding agents have "massively" sped up the team's work since Opus 5.5. Separately, the company showed Prime Sandboxes running 20,000 sandboxes at about 350 per second, though the snippet is cut off before the full details.
AI-assisted summary based on the listed source.
What happened
Prime Intellect's open-source reinforcement-learning trainer prime-rl 0.11.0 adds a new pretraining entry point, faster training with model-specific kernels, two-way KV-cache transfer and lower CPU memory use. Developer @samsja19 said coding agents have "massively" sped up the team's work since Opus 5.5....
Why it matters
Developers who train models with open-source reinforcement-learning tools may find this release relevant, since it targets training speed, memory use and cache handling. The claim about coding agents comes from the post's author and has not been independently checked here. Readers should look for the full sandbox throughput details before drawing conclusions about scale.
Signal Intelligence
Signal Strength 81%
Technical label WATCH
Public Interest 41
Category OPEN SOURCE
Reader Depth TECHNICAL
Signal Strength reflects source quality, relevance, freshness and evidence. Public Interest helps organize discovery; it is not proof of truth.
Public Interest components
Recognizable Entity Score 0
Practical Impact Score 46
Novelty Interest Score 94
Consequence Score 38
Curiosity Score 16
Shareability Score 53
Why this is here
VQV surfaced this signal because it is recent, relevant to AI Coding Tools, connected to Scobleizer News Pulse (X).