Understand this story
On October 20, 2026, an open-source environment model post-trained with reinforcement learning based on Qwen3.8-27B was released. According to Rohan Paul, the model reduced the hallucination rate on AA-Omniscience from 30.4% to 15.1%, and surpassed Muse Glimmer 30B and GPT-6 Luna high on Terminal-Bench, τ3-bench, ClawEval, and BFCL v4; however, on DeepSWE, Luna still leads with 53.1%. Its median output for solved tasks is 23.8K tokens, compared with 54.9K for the base Qwen. The authors also mention that the MLA version replaces 16 layers of full attention with multi-head latent attention, reducing memory usage per token from 64 KiB to 18 KiB; the compressed version can run on a single 24 GB RTX 4090, with a context about 3.5 times that of the base Qwen.
AI-generated from public reports
Developments
1 confirmed developments
- Source:Rohan Paul
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Oct 20
Rohan Paul
The Qwen3.8-27B reinforcement-learning post-trained model surpasses GPT-6 Luna high on multiple Agent benchmarks
Rohan Paul introduced an open-source environment model built on Qwen3.8-27B with reinforcement-learning post-training, which reduced the hallucination rate on AA-Omniscience from 30.4% to 15.1%. The model surpasses Muse Glimmer 30B and GPT-6 Luna high on Terminal-Bench, τ³-bench, ClawEval, and BFCL v4, but Luna still leads on DeepSWE at 53.1%; its median output for solved tasks is 23.8K tokens, compared with 54.9K for base Qwen. The author also mentions that the MLA version replaces 16 layers of full attention with multi-head latent attention, reducing memory usage per token from 64 KiB to 18 KiB; the compressed version can run on a single 24 GB RTX 4090, with a context about 3.5 times that of base Qwen.