--- title: SlopCodeBench 10/2026 – Opus 5.5, Astra wins overall,... description: New SlopCodeBench results show gpt-6 Astra topping frontier coding agents with a 16.3% strict pass rate, up from gpt-5.5's 14.8% in May. The test ran 36... canonical: https://daily.dev/posts/slopcodebench-10-2026-opus-5-5-astra-wins-overall-xhigh-considered-harmful-mjhhblw2w twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: SlopCodeBench 10/2026 – Opus 5.5, Astra wins overall, xhigh considered harmful | daily.dev og:description: New SlopCodeBench results show gpt-6 Astra topping frontier coding agents with a 16.3% strict pass rate, up from gpt-5.5's 14.8% in May. The test ran 36... og:url: https://daily.dev/posts/slopcodebench-10-2026-opus-5-5-astra-wins-overall-xhigh-considered-harmful-mjhhblw2w og:image: https://api.daily.dev/og/posts/mjhhBlW2w.png og:image:alt: SlopCodeBench 10/2026 – Opus 5.5, Astra wins overall, xhigh considered harmful 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. # SlopCodeBench 10/2026 – Opus 5.5, Astra wins overall, xhigh considered harmful **[Hacker News](https://daily.dev/sources/hn)** · 9 min read · 3 upvotes · 2 comments ## Summary New SlopCodeBench results show gpt-6 Astra topping frontier coding agents with a 16.3% strict pass rate, up from gpt-5.5's 14.8% in May. The test ran 36 problems and 196 checkpoints across Opus 5.5, Fable 5.1, SOL 6/6.1, Astra, and GLM 5.3, all scoring 12.2-16.3% strict. Higher 'XHigh' reasoning effort cost 2.11-4.22x more than Medium and was not a reliable way to get more strict passes, notably for Opus 5.5. Opus 5.5 Medium left the most leftover 'slop' code at 94.25% flagged lines, while Opus XHigh was cleanest at 77.75% despite fewer strict passes. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What strict pass rate did gpt-6 astra get on SlopCodeBench compared to gpt-5.5? Gpt-6 astra scored a 16.3% strict pass rate on SlopCodeBench, up from gpt-5.5's 14.8% recorded in May. Strict pass rate means a model passed every test case in a checkpoint including all regression tests. The October run covered 36 problems and 196 checkpoints, with frontier models including Opus 5.5, Fable 5.1, SOL 6/6.1, and GLM 5.3 scoring between 12.2% and 16.3% strict. _Developers tracking which coding agent to adopt can follow benchmark updates like this on daily.dev._ ### Does using higher reasoning effort (xhigh) improve coding agent benchmark scores? Not reliably. On SlopCodeBench, xhigh effort cost 2.11 to 4.22 times more than medium effort across three model pairs, but strict pass rates only improved for gpt-6 astra (29 to 32). Opus 5.5 actually scored worse on xhigh (25) than medium (28), despite xhigh producing cleaner code with a lower slop percentage of 77.75% versus 94.25% for medium. _Anyone tuning reasoning-effort settings for coding agents can weigh these cost and accuracy tradeoffs via daily.dev._ ### How much leftover junk or 'slop' code do AI coding agents leave in a codebase? A large share, measured as the percentage of source lines flagged by over 200 slop-detection linters for anti-patterns and repeated code. Opus 5.5 Medium had the highest slop presence at 94.25%, while Opus 5.5 XHigh had the lowest at 77.75%. Opus Medium also left 216,881 source lines in one run, including 45 scratch files totaling 41,265 lines. _Teams evaluating code cleanliness from AI agents can track findings like this on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 1 discussion and 99 comments across hackernews (as of 2026-10-21). **TL;DR:** Commenters largely accept the benchmark's headline findings (medium effort often beats xhigh, Opus leaves more slop) but spend much of the thread debating practical workarounds for messy model output and whether multi-agent orchestration setups are actually worth the complexity. **Sentiment:** 15% positive · 55% mixed · 30% skeptical **The case for** - Several developers confirm medium reasoning effort is often as good or better than xhigh/max in their own experience, saving cost and time. - Some find value in using one model (e.g. Astra) for exploration/planning and delegating implementation to a cheaper, more focused model. - Running generated code through a formatter and strict lint/complexity checks is reported to mitigate a lot of the 'slop' problem. **The pushback** - Multiple commenters report Astra/newer models frequently produce unreadable code, like dropping line breaks or 100-line functions with no formatting. - Skepticism that multi-agent 'LARPing' orchestration setups add real value over just using a single frontier model directly. - Some view the benchmark's precision (reporting results to several significant digits off only 36 problems) as spurious. - Concern that LLMs are structurally biased toward adding/defensive code rather than simplifying, which benchmarks like this may not fully capture. - Doubts that the sloppiness metric meaningfully captures code quality/readability versus being gameable fuzzy math. **By community** - hackernews (mixed): Discussion splits between practical tips for taming messy model output, debate over whether multi-agent orchestration is useful or self-indulgent, and some methodological skepticism about the benchmark itself. **Hottest debate:** Whether elaborate multi-agent/multi-model orchestration setups provide real benefits or are just an engineering form of self-indulgent 'LARPing' compared to simply using one frontier model well. **Open questions** - Is there a representative subset of the 36 problems that could be run more cheaply while still correlating with full results? - How would other models like Gemini 4 or GPT-6.1 Sol (Low) perform on the same benchmark? - Can benchmarks be designed to measure minimal-diff/refactor quality rather than just pass/fail and raw slop line counts? **Highlights** > Astra is some of the worst code I've seen in terms of slop, so the benchmark is kind of surprising to me. Astra gets things right, but I need Claude clean up its slop all the time. Many times I've cut down Astra changes from 1000+ lines to ~100 lines. > — [loveparade on hackernews · 1 comments](https://news.ycombinator.com/item?id=50175065) > IMO slop usually comes from overly defensive programming from LLMs. They’re loath to remove code and simplify, because unless we humans tell them some code / use case is unimportant, they have no reason to not care about it. So agents can make these overly complex solutions trying to make bulletproof implementations and prevent regressions. It requires a human to look at code, ask about complexity, and guide simplification by subtracting requirements implicit in the code. > — [softwaredoug on hackernews · 1 comments](https://news.ycombinator.com/item?id=50173236) > No "may be" about it. There is no benefit to having bad models write your code, and there's no benefit to "plan" mode. Plan mode is an outdated 2025 concept that should be removed from standard harnesses. There are 2 scenarios for dumb builder agents: 1. Your "planner" actually implements the project almost completely, finds all the gotchas, solves all of the emergent problems, and fully plans the solution. 2. Your "planner" takes a half-assed first attempt, solving none of the major problems that will arise during implementation In scenario 1 -- just pay cache output on a few more tokens to finish the project. No need to spend twenty minutes pumping a few million tokens through deepseek regardless of how cheap it is to do so. In scenario 2 -- good luck getting quality output I do think this "software team" LARPing is an engineering version of AI psychosis. You can get all of the little parts humming along, sending messages, and looking really cool. You can even convince yourself that all of these busy little bees actually make better product. But a frontier model working alone can produce better output in a tiny fraction of the time. Just build the thing. > — [criley2 on hackernews · 2 comments](https://news.ycombinator.com/item?id=50174800) > I fear OpenAI models are collapsing on themselves (idk if it's because of loop transformers or too much RL etc), which I didn't feel with 5.6 Sol at all. Need a 6.1 Sol nor Astra really feel like full blown legitimate models. More like autistic genius individuals who are partially brain dead. Which can sometimes work incredibly well when working in conjunction with another model. It's unbelievable how incredible gpt-6.1-sol even on medium reasoning is at reviewing Opus 5.5's work it will find incredibly weird, but legitimate, edge cases, yet if you actually use it as the main model, it just misses forest for trees so much. They are more like an organ which can do an incredible job at a specialized task that the brain cannot handle efficiently but the second you try to do brain's work with time, it all comes crashing down. I hope they are able to recover with future models. If you could could somehow combine how Opus approaches a task with the autistic brilliance of Astra/6.1, you would have an incredible model far exceeding today's models. > — [jeffy29 on hackernews](https://news.ycombinator.com/item?id=50176207) > Wouldn't it be better to enforce pass via the harness and then measure cost/quality since that's more what people are likely to do? > — [foota on hackernews](https://news.ycombinator.com/item?id=50172517) **Source threads** - [hackernews](https://news.ycombinator.com/item?id=50108717) · 245 points · 99 comments ## Community discussion Top comments from developers on daily.dev. **@kartiknvj** · 0 upvotes > Opus 5.5 going from 28 to 25 strict passes between Medium and XHigh, at 4.22x the cost, made me rethink defaulting to max effort on long refactors, even with one run per setting. The regression column stood out more: 2,206 of Opus Medium's 2,508 final failed tests are regressions. Will Part II show which checkpoint types those cluster on? **@bob72** · 0 upvotes > Anthropic is cooked. All above sonnet 4.6 is rubbish, 5 and above could not produce anything useful. Don't think any dev work require Opus or Fable. Always surprised when hear from someone that Sonnet is not enough for dev work. ## Similar posts on daily.dev - [advanced-context-engineering-for-coding-agents/benchmarking-opus-5-on-slop-code-bench.md at main · humanlayer/advanced-context-engineering-for-coding-agents](https://daily.dev/posts/advanced-context-engineering-for-coding-agents-benchmarking-opus-5-on-slop-code-bench-md-at-main-h-l5bkkqkom) · Hacker News · 2 upvotes · 0 comments - [What 9,050 AI-Written Scripts Taught Me About Slop](https://daily.dev/posts/what-9-050-ai-written-scripts-taught-me-about-slop-il5in6rbj) · DEV · 0 upvotes · 0 comments - [AI’s best coding agent fails 60% of the time — and the data backs it up](https://daily.dev/posts/ai-s-best-coding-agent-fails-60-of-the-time-and-the-data-backs-it-up-ziebak7ti) · The New Stack · 2 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: [#ai](https://daily.dev/tags/ai), [#ai-agents](https://daily.dev/tags/ai-agents), [#claude](https://daily.dev/tags/claude), [#gpt](https://daily.dev/tags/gpt) [View this post on daily.dev](https://daily.dev/posts/slopcodebench-10-2026-opus-5-5-astra-wins-overall-xhigh-considered-harmful-mjhhblw2w) ```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":"SlopCodeBench 10/2026 – Opus 5.5, Astra wins overall, xhigh considered harmful","url":"https://daily.dev/posts/slopcodebench-10-2026-opus-5-5-astra-wins-overall-xhigh-considered-harmful-mjhhblw2w","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/slopcodebench-10-2026-opus-5-5-astra-wins-overall-xhigh-considered-harmful-mjhhblw2w"},"datePublished":"2026-10-20T21:00:23.094Z","dateModified":"2026-10-21T08:06:41.423Z","description":"New SlopCodeBench results show gpt-6 Astra topping frontier coding agents with a 16.3% strict pass rate, up from gpt-5.5's 14.8% in May. The test ran 36...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/d7f572ab3164ffd4a3512aedcdbddd4b?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/d7f572ab3164ffd4a3512aedcdbddd4b?_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":2,"discussionUrl":"https://daily.dev/posts/slopcodebench-10-2026-opus-5-5-astra-wins-overall-xhigh-considered-harmful-mjhhblw2w","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":3},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":2}],"keywords":"ai,ai-agents,claude,gpt","timeRequired":"PT9M"} {"@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":"SlopCodeBench 10/2026 – Opus 5.5, Astra wins overall, xhigh considered harmful"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/slopcodebench-10-2026-opus-5-5-astra-wins-overall-xhigh-considered-harmful-mjhhblw2w","comment":[{"@type":"Comment","text":"Opus 5.5 going from 28 to 25 strict passes between Medium and XHigh, at 4.22x the cost, made me rethink defaulting to max effort on long refactors, even with one run per setting. The regression column stood out more: 2,206 of Opus Medium’s 2,508 final failed tests are regressions. Will Part II show which checkpoint types those cluster on?","datePublished":"2026-10-21T11:51:16.580Z","url":"https://daily.dev/posts/mjhhBlW2w#c-PPFod1Bcd","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"}},{"@type":"Comment","text":"Anthropic is cooked. All above sonnet 4.6 is rubbish, 5 and above could not produce anything useful. Don’t think any dev work require Opus or Fable. Always surprised when hear from someone that Sonnet is not enough for dev work.","datePublished":"2026-10-22T00:24:45.241Z","url":"https://daily.dev/posts/mjhhBlW2w#c-Nnww2itVq","author":{"@type":"Person","name":"Bob","url":"https://daily.dev/bob72","image":"https://avatars.githubusercontent.com/u/193962566?v=4"}}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/slopcodebench-10-2026-opus-5-5-astra-wins-overall-xhigh-considered-harmful-mjhhblw2w#faq","mainEntity":[{"@type":"Question","name":"What strict pass rate did gpt-6 astra get on SlopCodeBench compared to gpt-5.5?","acceptedAnswer":{"@type":"Answer","text":"Gpt-6 astra scored a 16.3% strict pass rate on SlopCodeBench, up from gpt-5.5's 14.8% recorded in May. Strict pass rate means a model passed every test case in a checkpoint including all regression tests. The October run covered 36 problems and 196 checkpoints, with frontier models including Opus 5.5, Fable 5.1, SOL 6/6.1, and GLM 5.3 scoring between 12.2% and 16.3% strict. Developers tracking which coding agent to adopt can follow benchmark updates like this on daily.dev."}},{"@type":"Question","name":"Does using higher reasoning effort (xhigh) improve coding agent benchmark scores?","acceptedAnswer":{"@type":"Answer","text":"Not reliably. On SlopCodeBench, xhigh effort cost 2.11 to 4.22 times more than medium effort across three model pairs, but strict pass rates only improved for gpt-6 astra (29 to 32). Opus 5.5 actually scored worse on xhigh (25) than medium (28), despite xhigh producing cleaner code with a lower slop percentage of 77.75% versus 94.25% for medium. Anyone tuning reasoning-effort settings for coding agents can weigh these cost and accuracy tradeoffs via daily.dev."}},{"@type":"Question","name":"How much leftover junk or 'slop' code do AI coding agents leave in a codebase?","acceptedAnswer":{"@type":"Answer","text":"A large share, measured as the percentage of source lines flagged by over 200 slop-detection linters for anti-patterns and repeated code. Opus 5.5 Medium had the highest slop presence at 94.25%, while Opus 5.5 XHigh had the lowest at 77.75%. Opus Medium also left 216,881 source lines in one run, including 45 scratch files totaling 41,265 lines. Teams evaluating code cleanliness from AI agents can track findings like this on daily.dev."}}]} ```