Rohan Paul (@rohanpaul_ai) · 10/20/2026, 22:08:25
RL Post-trained Qwen3.8-27B Halves Hallucinations and Beats GPT-6 Luna
An RL post-trained version of Qwen3.8-27B trained on open-source environments significantly improves agent performance, reducing hallucination rates on AA-Omniscience from 30.4% to 15.1% and outperforming GPT-6 Luna-medium on Terminal-Bench. The model features an MLA architecture that fits on a single RTX 4090 with 3.5x more context than the base model, while generating fewer output tokens per task, indicating higher 'intelligence density'.
SOURCE COVERAGEFull Original Report
Rohan Paul (@rohanpaul_ai) · 2026-10-20 14
Some of the most interesting agent results this year are coming out of post-trains of small open models.
A new RL post-train Qwen3.8-27B model trained with an open-source environments cuts the hallucination rate on AA-Omniscience from 30.4% to 15.1%.
the compressed copy of the mla version fits on my single 24 GB RTX 4090 and holds about 3.5x more context than base Qwen.
It beats GPT 6 Luna-medium on DeepSWE, placing in between Luna 6 medium and high. scores higher on Terminal-Bench 2.1 while writing a median 23.8K output tokens per solved task, where base Qwen writes 54.9K.
Tried an unreleased version of this model, and really felt the comparative intelligent density, that it writes a lot less.
tops Muse Glimmer 30B and OpenAI’s GPT-6 Luna (high) on Terminal-Bench, τ³-bench, ClawEval and BFCL v4 (Berkeley Function Calling Leaderboard), though Luna still wins DeepSWE at 53.1%.
They also released an MLA version, which is the same model with its 16 full-attention layers converted to multi-head latent attention, which shrinks the memory each token takes from 64 KiB to 18 KiB so far more or longer sessions fit on one GPU.