researchHugging Face

Hugging Face releases Qwen3.8-27B-pi-adapted LoRA, lifting SWE-bench Pro hard-51 pass@1 from 39.2% to 64.7%

TL;DR

Hugging Face released Qwen3.8-27B-pi-adapted, an Apache-2.0 LoRA adapter trained with reinforcement learning for the pi coding agent. According to MarkTechPost, it raises SWE-bench Pro hard-51 pass@1 from 39.2% to 64.7% and Terminal-Bench 2.1 from 39.5% to 55.8%.

1 min read
0

Hugging Face released Qwen3.8-27B-pi-adapted, an Apache-2.0 LoRA adapter trained with reinforcement learning for the pi coding agent. MarkTechPost reported the release on November 18, 2026.

Reported results

According to MarkTechPost, the adapter produces the following gains over the base Qwen3.8-27B:

Benchmark Metric Base With adapter
SWE-bench Pro hard-51 pass@1 39.2% 64.7%
Terminal-Bench 2.1 score 39.5% 55.8%

Release details

  • Name: Qwen3.8-27B-pi-adapted
  • Publisher: Hugging Face
  • Type: LoRA adapter
  • Training method: reinforcement learning
  • Target: the pi coding agent
  • License: Apache-2.0

What this means

The release pairs a Qwen3.8-27B base with a LoRA adapter trained with reinforcement learning for one coding agent, pi. The reported gains on SWE-bench Pro hard-51 and Terminal-Bench 2.1 come from that agent-specific training. The Apache-2.0 license governs use of the adapter. Teams evaluating it should note that the reported results are tied to the pi agent setup.

Related Articles

research

Microsoft Research's TokenEconomizer plugin cuts coding agent costs 18.4% to 30.9% on SWE-Bench

TokenEconomizer, an open-source plugin from UC Berkeley, Microsoft Research and Microsoft Security, learns from an agent's past sessions in a repository to cut token costs. Across 277 SWE-Bench tasks and five agent and model combinations, mean cost per task fell 18.4% to 30.9%, according to the authors.

research

Amazon releases Mitra-v2, an open tabular foundation model that leads the single-model field on TabArena

Amazon has released Mitra-v2, a tabular foundation model for classification and regression that, according to Amazon, leads the single-model field on the TabArena benchmark of 51 datasets. Weights and code are available under Apache-2.0 on Hugging Face.

research

Google's Tunix hits bitwise-identical sampler and trainer logprobs for on-policy RL on 64 TPU v5p chips

Google engineers report a zero train-inference mismatch (Zero-TIM) implementation in Tunix, in which sampler and trainer assign bitwise-identical logprobs to the same token at the same weights. It was verified at multi-host scale across 64 TPU v5p chips (DP8×TP8). The recipe combines fixed-order Pallas kernels, explicit cross-shard reductions, a shared logprob path and custom differentiation rules.

research

NVIDIA recipe: RL post-training lifts Nemotron 3.5 Super VL accuracy a further 17% on drug-metabolism task

NVIDIA has published a recipe for post-training Nemotron 3.5 Super VL on scientific tasks using NeMo Gym and NeMo RL. In a drug-metabolism example, supervised fine-tuning improved accuracy by 12% and RL post-training lifted it a further 17% at a 16,384-token output budget.

Comments

Loading...