--- title: Five Qwen3.8-27B models match Fable 5 Using GVS5H Harness description: A training-free orchestration method lets multiple instances of the same model coordinate on hard coding problems through a shared filesystem plan and notes.... canonical: https://daily.dev/posts/five-qwen3-8-27b-models-match-fable-5-using-gvs5h-harness-qhkrpg47j twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Five Qwen3.8-27B models match Fable 5 Using GVS5H Harness | daily.dev og:description: A training-free orchestration method lets multiple instances of the same model coordinate on hard coding problems through a shared filesystem plan and notes.... og:url: https://daily.dev/posts/five-qwen3-8-27b-models-match-fable-5-using-gvs5h-harness-qhkrpg47j og:image: https://api.daily.dev/og/posts/qhkRpG47J.png og:image:alt: Five Qwen3.8-27B models match Fable 5 Using GVS5H Harness og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Five Qwen3.8-27B models match Fable 5 Using GVS5H Harness **[Hacker News](https://daily.dev/sources/hn)** · 3 min read · 4 upvotes · 1 comments ## Summary A training-free orchestration method lets multiple instances of the same model coordinate on hard coding problems through a shared filesystem plan and notes. On the 100 latest hard LiveCodeBench problems, orchestrated Qwen3.8 Flash Next hits 93.0% pass@1, beating Claude Fable 5's 90.4% at 9% of the cost. Open-weight Qwen3.8-27B rises from 66.8% to 92.4% when orchestrated. Gains reach up to 25.6 percentage points on pinned backends, though some models see no improvement or regress. The code runs via uv with a multiagent or single-call engine flag. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Can orchestrating multiple instances of a smaller open-weight model match a frontier closed model on coding benchmarks? Yes, orchestrated open-weight Qwen3.8-27B rises from 66.8% to 92.4% pass@1 on the 100 latest hard LiveCodeBench problems, while orchestrated Qwen3.8 Flash Next reaches 93.0%, surpassing Claude Fable 5's 90.4% single-call score at roughly 9% of the cost. The method uses a shared filesystem holding a plan, notes and current solution for coordination. _Comparing orchestration strategies against frontier models is easier with benchmark write-ups like this tracked on daily.dev._ ### What is ledger-based zero-shot self-orchestration in multi-agent LLM coding? It is a training-free method where fresh instances of one model decompose a coding problem and coordinate through a shared filesystem that holds a plan, notes and the current solution. Tested across eleven open and closed-weight models on 100 hard LiveCodeBench problems, it produced gains of up to 25.6 percentage points on pinned backends, though some models showed no improvement or got worse. _Developers evaluating multi-agent coding setups can follow emerging orchestration techniques like this on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 4 discussions and 77 comments across hackernews (as of 2026-10-25). **TL;DR:** Discussion splits between practical interest in running Qwen3.8-27B locally with various harnesses, and sharp skepticism about whether the benchmark results (an orchestrated model matching a frontier model) are methodologically sound or just overfitting/benchmark gaming on a stale, likely-contaminated dataset. **Sentiment:** 20% positive · 35% mixed · 45% skeptical **The case for** - Several people confirm Qwen3.8-27B is genuinely capable and already use similar shared-file/notes workflows for agentic coding without a formal harness. - Running it locally on consumer GPUs (e.g. RX 7900 XTX) is seen as cost-effective after the one-time hardware cost. - Cerebras and other fast-inference backends show the orchestrated approach can run quickly despite running multiple instances. **The pushback** - Several argue the benchmark is an old, likely-contaminated dataset (LiveCodeBench problems through April 2025) so the comparison is not meaningful. - Commenters say there's no comparison against other harnessed/orchestrated frontier models, only against a single-pass baseline, making the claim of beating Fable 5 misleading. - Multiple people suspect the harness is essentially overfitting to or eval-maxxing the specific benchmark rather than reflecting real agentic ability. - One commenter calls out that ignoring the benchmark's intended single-shot rules while adding tools/agentic loops breaks the validity of the comparison. **By community** - hackernews (mixed): Threads diverge into a long practical sub-discussion about local harnesses/hardware for running Qwen3.8-27B, alongside a pointed critical thread questioning the benchmark methodology and contamination risk. **Hottest debate:** Whether the headline benchmark result reflects genuine capability gains from orchestration or is just overfitting/eval-gaming on a stale, likely-contaminated dataset without comparison to other harnessed frontier models. **Open questions** - Would the orchestration approach still beat Fable 5 on a fresh, uncontaminated benchmark or on real agentic-work benchmarks? - How does orchestrated Qwen3.8 compare to similarly-harnessed versions of other frontier models like Astra/6.1 Sol/Fable 5.1/Opus 5.5? - Is the single-worker-with-fresh-context design actually necessary, or could a simpler shared-notes approach achieve the same gains? **Highlights** > So using this harness five qwen3.8-27b models roughly matched Fable 5 on a saturated coding benchmark from early 2025. No comparison to astra/6.1 sol/fable 5.1/opus 5.5. No comparison on a benchmark that reflects actual agentic work (and agentic behavior is what modern models optimize for anyway - probably makes qwen look worse than it is?). And also obviously using a very old benchmark like this: > The updated release of the dataset with problems released between May 2023 and Apr 2025 containing 1055 problems means essentially guarenteed contamination. I have no idea if this harness is any good. Agressive orchestration could be interesting given how hard models are RLed on orchestration? But the methods used to evaluate this harness are not informative > — [vikramkr on hackernews · 2 comments](https://news.ycombinator.com/item?id=50228389) > … on one benchmark, that has 2 modes of running. 1 shot and 1 shot with repair. Zero tools. Then ignore the rules of the bench (which is a pretty obsolete one anyway), give it an agentic loop of sorts via delegation and python eyes and then win. Sorry but this is not how benchmarking works > — [sams99 on hackernews](https://news.ycombinator.com/item?id=50228420) > Harness changes dynamically depending on your LLM intelligence. The smarter LLMs get, you apply less, as LLM takes care of pre-existing harnesses. This is a shitty, unreliable, and unmaintainable way of using harness because the harness is built around for the sake of getting good benchmark scores (someone mentioned overfitting in comments, so that) > — [yipinwong on hackernews](https://news.ycombinator.com/item?id=50228542) > Interesting but (from the PDF): Every role but the verifier is the same underlying model, called in a fresh context and coordinating through a shared workspace on disk: /task.md - the problem statement /plan.md - the manager’s overarching plan /tasks.json - a debug mirror of the task list /notes.md - curated ideas / findings / partial proofs /solution.py - the current solution Then they describe how there is a ping-pong between manager and worker with fresh context for the worker, but i wonder, is that part or even the worker really necessary? My use of Qwen 3.8 27B (which is the only model i use almost since it came out) is really doing something very similar to the above where first a task is described, a `PLAN.md` file is made with the (current) plan and tasks are either in the plan file or in a separate issue tracker the LLM has access to via a CLI tool, a `STATUS.md` that tracks the current status and a `GOTCHAS.md` file for writing down various notes/tidbits/gotchas (TBH this is largely because Qwen 3.8 27B does want to note down stuff anyway so it mainly acts as an attractor for not writing its notes down in the other two places :-P). There isn't any real harness driving this, i just give it a task, tell it how it'll be handled and off it goes. I guess i'm playing the "role" of the harness there as i tell it to not go and do everything in one go :-P but i never bother with fresh contexts, Pi's "rolling" approach to compaction (i.e. summarizing the older messages to make space while keeping the recent messages as-is) make it almost a non-issue. I have considered making it a planning tool, though as some tool it can use via the CLI (it is running in a container), not a harness - but at the same time it feels like it can do the same thing with the `PLAN.md`/`STATUS.md`/`GOTCHAS.md` triplet. The only thing the tool would help with is to provide a UI for me to check what it's currently doing, but even then i usually just stop it and ask it directly :-P then tell it to continue. Perhaps at some point i'll make some custom benchmark with stuff that are vaguely similar to what i work on and try a bunch of different approaches. It is something i want to do anyway to see how and if Qwen4 27B (whenever that comes it) is an improvement over 3.8. > — [badsectoracula on hackernews · 3 comments](https://news.ycombinator.com/item?id=50228427) > the qwen flash next result is the actual interesting part. The 27b dense model has more active parameters than, e.g. latest deepseek flash. > — [112233 on hackernews · 1 comments](https://news.ycombinator.com/item?id=50228399) **Source threads** - [hackernews](https://news.ycombinator.com/item?id=50228167) · 130 points · 75 comments - [hackernews](https://news.ycombinator.com/item?id=49671996) · 6 points · 0 comments - [hackernews](https://news.ycombinator.com/item?id=49730257) · 5 points · 2 comments - [hackernews](https://news.ycombinator.com/item?id=49740680) · 2 points · 0 comments ## Community discussion Top comments from developers on daily.dev. **@kartiknvj** · 0 upvotes > 66.8% to 92.4% on Qwen3.8-27B with nothing but fresh instances sharing a plan, notes and a solution file is a big jump for zero training. LCB 3765 is the case I keep thinking about, where the brainstorm talked itself out of the optimization the single call had used. Since every worker inherits the notes, I'm curious how much the fresh-perspective workers from the paper would pull that back. ## Similar posts on daily.dev - [Qwen3.8-Max Just Passed Claude Fable 5 on the Frontend Leaderboard. We Compared Them on 10 UIs](https://daily.dev/posts/qwen3-8-max-just-passed-claude-fable-5-on-the-frontend-leaderboard-we-compared-them-on-10-uis-hwjkn2zkh) · Kilo Blog · 1 upvotes · 0 comments - [The Agentic Analytics Benchmark, round two: GPT-6, Claude Opus 5.5, Claude Sonnet 5.5](https://daily.dev/posts/the-agentic-analytics-benchmark-round-two-gpt-6-claude-opus-5-5-claude-sonnet-5-5-l8whcfkkx) · ClickHouse · 1 upvotes · 0 comments --- Tags: [#python](https://daily.dev/tags/python), [#ai-agents](https://daily.dev/tags/ai-agents), [#qwen](https://daily.dev/tags/qwen) [View this post on daily.dev](https://daily.dev/posts/five-qwen3-8-27b-models-match-fable-5-using-gvs5h-harness-qhkrpg47j) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Five Qwen3.8-27B models match Fable 5 Using GVS5H Harness","url":"https://daily.dev/posts/five-qwen3-8-27b-models-match-fable-5-using-gvs5h-harness-qhkrpg47j","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/five-qwen3-8-27b-models-match-fable-5-using-gvs5h-harness-qhkrpg47j"},"datePublished":"2026-10-25T05:20:03.707Z","dateModified":"2026-10-25T07:20:53.802Z","description":"A training-free orchestration method lets multiple instances of the same model coordinate on hard coding problems through a shared filesystem plan and notes....","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/1797007ee2909202552f196182e4cd3e?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/1797007ee2909202552f196182e4cd3e?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Hacker News","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Hacker News","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/hn","url":"https://daily.dev/sources/hn"},"commentCount":1,"discussionUrl":"https://daily.dev/posts/five-qwen3-8-27b-models-match-fable-5-using-gvs5h-harness-qhkrpg47j","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":4},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":1}],"keywords":"python,ai-agents,qwen","timeRequired":"PT3M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Hacker News","item":"https://daily.dev/sources/hn"},{"@type":"ListItem","position":3,"name":"Five Qwen3.8-27B models match Fable 5 Using GVS5H Harness"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/five-qwen3-8-27b-models-match-fable-5-using-gvs5h-harness-qhkrpg47j","comment":[{"@type":"Comment","text":"66.8% to 92.4% on Qwen3.8-27B with nothing but fresh instances sharing a plan, notes and a solution file is a big jump for zero training. LCB 3765 is the case I keep thinking about, where the brainstorm talked itself out of the optimization the single call had used. Since every worker inherits the notes, I’m curious how much the fresh-perspective workers from the paper would pull that back.","datePublished":"2026-10-26T11:52:11.239Z","url":"https://daily.dev/posts/qhkRpG47J#c-IKnEIj1Az","author":{"@type":"Person","name":"kartik-nvjk","url":"https://daily.dev/kartiknvj","image":"https://media.daily.dev/image/upload/s--3gGgsVCw--/f_auto/v1781456774/avatars/avatar_TvTVeiMdkRCqWUDullFmy?_a=BAMAMiWQ0"}}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/five-qwen3-8-27b-models-match-fable-5-using-gvs5h-harness-qhkrpg47j#faq","mainEntity":[{"@type":"Question","name":"Can orchestrating multiple instances of a smaller open-weight model match a frontier closed model on coding benchmarks?","acceptedAnswer":{"@type":"Answer","text":"Yes, orchestrated open-weight Qwen3.8-27B rises from 66.8% to 92.4% pass@1 on the 100 latest hard LiveCodeBench problems, while orchestrated Qwen3.8 Flash Next reaches 93.0%, surpassing Claude Fable 5's 90.4% single-call score at roughly 9% of the cost. The method uses a shared filesystem holding a plan, notes and current solution for coordination. Comparing orchestration strategies against frontier models is easier with benchmark write-ups like this tracked on daily.dev."}},{"@type":"Question","name":"What is ledger-based zero-shot self-orchestration in multi-agent LLM coding?","acceptedAnswer":{"@type":"Answer","text":"It is a training-free method where fresh instances of one model decompose a coding problem and coordinate through a shared filesystem that holds a plan, notes and the current solution. Tested across eleven open and closed-weight models on 100 hard LiveCodeBench problems, it produced gains of up to 25.6 percentage points on pinned backends, though some models showed no improvement or got worse. Developers evaluating multi-agent coding setups can follow emerging orchestration techniques like this on daily.dev."}}]} ```