Prime Intellect releases prime-rl v0.10.0 with IPO loss as default

Prime Intellect releases prime-rl v0.10.0 and verifiers v0.4.0, making its new RL algorithm IPO the default training loss.

IPO replaces the previous DPPO-based default, which the release removes. The company says IPO is tailored for async RL stability and that the release adds performance improvements and new model support.

The first-party blog post reports an agentic SWE RL ablation on GLM-4.5-Air. In that test, IcePop collapsed around step 1.2k while IPO at eps 0.3 trained through, masking about 0.25% of tokens versus about 2% at the old 0.1. The post also says the release adds a single uv run eval command for training and standalone evaluation. The company's X post does not include these benchmark figures, and neither source names the newly supported models.