Tag
open-weight model evaluation
1 published trend · vote, save or discuss any of them
- 01AI for BusinessEmerging7D: Insufficient history
DaisyRose-27B: RadixArk's Agent Post-Training Recipe
RadixArk says its DaisyRose-27B combines failure-driven task curation, asynchronous reinforcement learning, long-session summarization and disposable sandboxes to post-train an agent model. The Apache 2.0-licensed checkpoints are inspectable, but the benchmark and efficiency figures come from the developer and have not been independently validated.
Published
AI agents & automation50 out of 100- 7D
- +100%
- Signals
- 1
- Indep. src
- 1
- Confidence
- Unconfirmed