AI News Daily한국어
Friday, November 20, 2026

Story 4 of 13

A 300MB adapter lifts a local 27B coding model from 35% to 51% on SWE-bench Pro, beating max reasoning with thinking off

First seen on X 2 days ago@TeksEdge ♥ 1,61715 views

A 300MB adapter lifts a local 27B coding model from 35% to 51% on SWE-bench Pro, beating max reasoning with thinking off

A post about a 300MB add-on that makes a local 27B coding model much better drew 1,600 likes and 2,100 bookmarks on X. It's Qwen3.8-27B-opencode-adapted, released Nov. 18 by the Harness Hub team on Hugging Face.

According to the model card, it's a LoRA adapter for Alibaba's Qwen3.8-27B, trained with reinforcement learning to run behind the open-source coding agent OpenCode. On a hard 51-task subset of SWE-bench Pro it goes from 35.3% to 51.0% (+15.7 points), and on Terminal-Bench 2.1 from 46.4% to 52.4%. On tasks both versions solve, median turns drop from 48 to 13. With thinking off, the adapter beats the base model at its maximum reasoning effort on SWE-bench Pro (51.0% vs 47.1%) while generating six to eight times fewer tokens, the team says.

It's Apache 2.0 and ships with instructions for vLLM, transformers and llama.cpp, plus GGUF files, so it runs on a home GPU. The card recommends leaving thinking off since turning it on adds nothing significant, and the team says it will publish the full training recipe so anyone can build adapters for other harnesses.

The caveat: every number comes from the OpenCode harness with one attempt per task and a 100-step budget, so the team says it isn't comparable to the official Qwen card, which uses other harnesses such as Claude Code and the full SWE-bench Pro. One reply said weight adapters "make your local models of any type dramatically better."

Comments 0

Select a sentence in the story (long-press on phones) to comment on it.

“So now a small 300MB adapter makes your local 27B coding model dramatically better!”

Next story →Ubisoft cut 166 more jobs the same day it unveiled Motherbrain, a generative AI studioRead an AI Insight →The generation that starts as team leads: KAIST's Prof. Minhyuk Sung says "'GPT did it' is no excuse"

Get tomorrow's stories as an alert

We'll ping you when each new edition or breaking story lands. Free, no sign-up.

Daily morning AI news, new AI Insights, and a monthly briefing of top stories on the 1st.

AdPut your product here. About 5,000 people a month read AI news on this site.Advertise with us ›

More stories from the same day

  1. Investors sold AI stocks to buy Anthropic: ANT closes up 17% on day one as the Nasdaq drops 2.25%
  2. OpenAI researchers would give up 40% of their team to keep coding agents; the top 10% burn $23,000 a day
  3. Apple is building its own computer-use AI agent, and made rivals' agents take 14 setup steps instead of four
  4. Ubisoft cut 166 more jobs the same day it unveiled Motherbrain, a generative AI studio
  5. Google will launch a data-center rack's worth of TPUs into orbit on Blue Origin's New Glenn in 2029
  6. "14 minutes. Never seen anything that fast": ex-NSA hacker says an AI agent swarm beat 2FA and encrypted a whole company
  7. A researcher sent 2 million bait prompts to AI services. Three providers turned out to be already breached
  8. Nearly 1,000 AI agents left together for three months wrote 936 papers and invented peer review
  9. Anthropic mathematician says he and Claude proved the Riemann hypothesis; no one has verified it yet
  10. SpaceX borrows $44 billion for Nvidia chips in the largest GPU financing yet, at yields above 8%
  11. Anthropic falls to within a nickel of its IPO price on second trading day
  12. SEC is probing Nvidia's accounting, Reuters reports; shares fall about 10%
See all of Friday, November 20, 2026 →

* Summaries and publishing on this page are done by an automated agent. Check the facts against the original links.

Privacy PolicySupportSend feedbackAdvertiseDiscord botClaude Code & CodexRSSListed on Sell With boost