Qwen3.8-27B RL后训练版：幻觉率减半，单卡4090可跑
一个 27B 开源模型把幻觉率砍半，单张 4090 就能跑，跑分还压过 GPT-6 Luna，这个后训练思路很值得看看。
基于 Qwen3.8-27B 的开源 RL 后训练模型将 AA-Omniscience 上的幻觉率从 30.4% 降到 15.1%。其 MLA 版本把每个 token 的显存占用从 64 KiB 压到 18 KiB，压缩后可在单张 24 GB RTX 4090 上运行，上下文容量约为原版 Qwen 的 3.5 倍。在 Terminal-Bench、τ³-bench、ClawEval 和 BFCL v4 上超过 Muse Glimmer 30B 和 GPT-6 Luna（high）。解决每个任务的中位输出 token 从基座 Qwen 的 54.9K 降到 23.8K，但 DeepSWE 上仍落后于 Luna（53.1%）。
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.
