# A small file took an open model from 35% to 51% on hard coding tasks. What a LoRA is, and what it changes outside code (slim)

> Vlad Cealicu on LoRA adapters. In November 2026 a 0.6 GB adapter took Qwen3.8-27B from 35.3% to 51.0% on the hardest tasks of SWE-bench Pro without changing the model. The same idea is how video models get faster, how a phone runs many features on one model and how a robot learns a new job, and in 2023 it was how the safety training of an open model was undone for under $200.

- Full: https://www.anchorterminal.com/blog/what-a-lora-is-and-what-it-changes.md (~3,100 tokens) · this version ~1,280 tokens · JSON https://www.anchorterminal.com/blog/what-a-lora-is-and-what-it-changes.json · canonical https://www.anchorterminal.com/blog/what-a-lora-is-and-what-it-changes
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-11-22
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)

By Vlad Cealicu*, 22 November 2026. 1,337 words, about 6 min to read.

- 35.3% to 51.0%: Qwen3.8-27B on the 51 hardest tasks of SWE-bench Pro, without and with the adapter, thinking off
- 6 to 8 times: fewer tokens than the plain model needs with thinking on, for about the same accuracy
- 10,000 times: fewer parameters to train than full fine-tuning, for GPT-3 in the paper that introduced LoRA
- 10s of megabytes: one adapter for Apple's on-device model of about 3 billion parameters

harness-hub's card for its OpenCode adapter [1], Hu et al.

In November 2026 an organisation on Hugging Face called harness-hub published a file that makes Qwen3.8-27B better at working inside the OpenCode coding agent [[1]](#ref-1).

That file is a LoRA adapter.

## What a LoRA is

A model is a very large set of numbers, and fine-tuning used to mean changing all of them, which gives you a second copy of the whole model for every job you train it for.

What comes out is a file that holds only those matrices.

## Why video and images got there first

An image or video model is expensive to retrain and slow to run as it comes, and a LoRA helps with both.

So in video an adapter is how you get what you want and how you get it quickly, and the base model is the thing everybody shares.

## What changed for code

The harness-hub adapters are trained for a harness, which is the coding agent a model works inside.

The accuracy went up, and I'd hold that loosely.

The number I'd hold on to is the work.

## What it changes

A lab isn't the only one who can make an open model better any more.

The harness is now part of the model.

One base model can do many jobs on a small machine.

It reaches robots too.

The same door opens for people you'd rather it didn't.

And the base model becomes hard to leave.

## What we're doing with it

Our lab's harness isn't OpenCode.

## What isn't known

Whether any of this holds outside the publisher's own tests.

---

More in [Opinion](https://www.anchorterminal.com/blog/opinion/index.md) · Older opinion [A machine that proved the Riemann hypothesis, or at least we think it did](https://www.anchorterminal.com/blog/a-machine-that-proves-riemann.md)

- Older: [An open model on a £3,879 box scored 98 on our test. Claude scored 100](https://www.anchorterminal.com/blog/local-models-on-one-box-against-claude.md)
- All posts: https://www.anchorterminal.com/blog/index.md

Anchor Terminal grades the tools AI agents run on, with a source for every score.

## References

Read on 22 November 2026.

[1] harness-hub, Qwen3.8-27B-opencode-adapted, model card on Hugging Face, November 2026. Source for the adapter's results, their intervals, the turns and tokens a task, its rank, its sizes, the size of its file (and of the GGUF build in harness-hub/Qwen3.8-27B-opencode-adapted-GGUF), that it needs its base model, that its numbers aren't comparable with Qwen's card and that its GGUF build hasn't been evaluated again.

[2] Hugging Face, Introducing harness adapters, 20 November 2026. Source for one adapter for each harness, trained inside it on two GPUs, the three harnesses, and what happened when the three adapters were merged.

[3] Hu and others, LoRA: Low-Rank Adaptation of Large Language Models, arXiv 2106.09685, 17 June 2021. Source for what a LoRA is, 10,000 times fewer trainable parameters, a third of the GPU memory and no added latency.

[4] Luo and others, LCM-LoRA: A Universal Stable-Diffusion Acceleration Module, arXiv 2311.05556, 9 November 2023. Source for a speed-up stored as a LoRA and plugged into other fine-tuned models without training.

[5] lightx2v, Wan2.2 Distilled LoRA Models, model card on Hugging Face.

[6] Apple Machine Learning Research, Introducing Apple's On-Device and Server Foundation Models, 10 June 2024. Source for one on-device model of about 3 billion parameters with an adapter for each feature, tens of megabytes each, loaded and swapped.

[7] ggml-org, llama.cpp, the server's README.

[8] Kim and others, OpenVLA: An Open-Source Vision-Language-Action Model, arXiv 2406.09246, 13 June 2024. Source for a 7 billion parameter robot model fine-tuned on consumer GPUs with low-rank adaptation.

[9] Anchor Terminal, anchorlab, MODELS.md and README.md.

[10] Anchor Terminal, Lab score, open-weight models on our own evaluation.

[11] Lermen and others, LoRA Fine-tuning Efficiently Undoes Safety Training in Llama 2-Chat 70B, arXiv 2310.20624, 31 October 2023. Source for under $200, one GPU and refusals of about 1%.

[12] Anchor Terminal, anchorlab, pull request 36. Source for the adapter and the plain model at its settings going on the lab's list.
