specific method · filed under post-training
Reinforcement learning on synthetic agentic data
Large-scale reinforcement learning conducted on synthetic agentic tasks, with the evidence specifying non-thinking mode.
Also called large-scale RL on synthetic data.
- source
- 1
- model
- 1
- lab adopt it
- 1
- strongest
- used
How sources treat it
One count per evidence span, weakest treatment to strongest.
evaluated 1used 1
Documented in
Evidence
2 spans quoted from the sources, strongest treatment first.
large-scale RL on synthetic data yields substantial improvements over DeepSeek-V3.2-SFT on Tau2Bench, MCP-Mark, and MCP-Universe benchmarks.
usedoptimizationin DeepSeek-V3.2-SFTDeepSeek
Filed alongside
Other methods under post-training :: agentic post-training.
Agentic reinforcement learningMulti-harness trainingAgentic tool-use trainingEnvironment hardeningRepair-agent environment correction loopAgentic post-trainingAutonomous Execution Tasks (AET)Concurrent multi-environment post-trainingContainer-level network isolationEnvironment preparation to prevent solution leakageEnvironment scaling for realistic training tasksGenerate-verify-refine loopHarness-optimized trainingMock applications for personal-assistant reinforcement learningMulti-agent task decomposition and coordinationMulti-environment reinforcement learningMulti-environment reinforcement learning from verifiable rewardsMulti-harness trajectory training with native behavior preservationMulti-turn collaboration simulatorProduction-harness-matched reinforcement learningPython tool use in chain-of-thoughtRe-post-training for agentic capabilitiesReinforcement learning in an agent harnessReinforcement learning restricted to search and code environments