{"slug": "a-cheapskate-s-guide-to-coding-with-expensive-models", "title": "A cheapskate's guide to coding with expensive models", "summary": "Letting Opus drive cheaper Luna subagents through the Mjolnir harness cuts coding spend by about 60% with no loss of intelligence, at the cost of roughly doubled task latency, according to a SlopCop analysis by Jonathan Ellis. Opus+Luna is also cheaper and smarter than Sol, but GPT models burn their token savings busywaiting on subagent results, while Claude models in Claude Code perform best. SlopCop also reports that an Oracle pattern, in which weaker DeepSeek Flash consults a stronger advisor model, moves Flash about one-third of the way to Opus in both intelligence and speed while doubling price.", "body_md": "- Written by\n- [Jonathan Ellis](https://blog.slopcop.com/author/jonathan/)\n- Published\n- By email\n- [Subscribe](#/portal/signup)\n\nIt seems straightforward that you should be able to save money by having a smart, expensive model delegate straightforward tasks to cheaper models, and this is in fact the case. With [Mjolnir](https://github.com/BrokkAi/mjolnir/?ref=blog.slopcop.com), you can even do this cross-provider; here's price vs performance of Opus solving uncontaminated Rust tickets with and without Luna/max subagents. (Luna is of course [far cheaper than Haiku](https://slopcop.com/power-ranking?hidden=deepseek-flash-dsh%2Fhigh%2Cdeepseek-flash-dsh%2Fmax%2Cglm-5.3-flash%2Fmax%2Cglm-5.3%2Fmax%2Cgpt-5.6-luna%2Fmax%2Cgpt-5.6-sol%2Fxhigh%2Cgpt-6-astra%2Flow%2Cgpt-6-astra%2Fmedium%2Cgpt-6-sol%2Fxhigh%2Cgpt-6.1-sol%2Fhigh%2Cgpt-6.1-sol%2Fmedium%2Cgpt-6.1-sol%2Fxhigh%2Cgpt-6.1-terra%2Fhigh%2Cgpt-6.1-terra%2Fmax%2Cgpt-6.1-terra%2Fxhigh%2Cgrok-4.6%2Fxhigh%2Cgrok-4.7%2Fhigh%2Cgrok-4.7%2Fxhigh%2Ckimi-k3.1-pro%2Fhigh%2Ckimi-k3.1-pro%2Fmax%2Ckimi-k3%2Fmax%2Cmuse-spark-1.3%2Fultra%2Copus-5.5%2Fhigh%2Copus-5.5%2Fmedium%2Copus-5.5%2Fxhigh%2Copus%2Fxhigh%2Csonnet-5.5%2Fhigh%2Csonnet-5.5%2Fmedium%2Csonnet-5.5%2Fxhigh&ref=blog.slopcop.com).) Not only is Opus+Luna cheaper than Opus, it's also cheaper (and smarter) than Sol:\n\n### The fine print\n\nYou can use ~any model effectively as a subagent model, although you'll obviously want to choose the ones with the best [price:performance](https://slopcop.com/power-ranking?ref=blog.slopcop.com). But not every model and not every harness can drive subagents effectively. In particular, GPT models will burn all of their token savings by busywaiting on subagents' results, even though Mjolnir's subagent-management tools are designed to discourage this. Claude models running in Claude Code (which has the extremely effective `monitor` tool as an alternative to polling) do best.\n\n### Beyond one subagent\n\nAlthough the confidence interval is wide, it does look like Luna makes enough mistakes that it can degrade Opus's performance. And for trivial tasks, Luna/max wastes a lot of tokens (and more importantly, time) by overthinking.\n\nAt first we tried to solve this with an adaptive difficulty level, asking Jev to classify what effort to run Luna at. But we couldn't figure out how to give Jev effective anchors, or maybe Jev just isn't smart enough; in any case out of 56 tasks, Jev classified 50 of them as xhigh.\n\nSo instead we split Mjolnir's subagents up into \"straightforward\" and \"subtle\" configurations, and iterated on the prompting until Claude understood how to use them effectively. This let us dial down straightforward tasks to Luna/high instead of max, while using a smarter model (DeepSeek Flash/max on API, or Sol/medium on subscription) for trickier ones.\n\n### Inversion of control\n\n\"But wait,\" you ask. \"What is the Oracle setting?\"\n\nI'm glad you asked!\n\nThe Oracle allows you to configure a smarter and more expensive advisor model that a weaker primary model can consult as needed.\n\nThis is also useful, but a bit more of a mixed bag. Here's how it stacks up:\n\nHere, \"Flash subagents\" are configured with low effort for straightforward, and max for subtle. All the opuses run at xhigh effort.\n\n## Conclusion\n\nYou can save about 60% of your spend without giving up any intelligence by letting the best coding model in the world drive cheaper subagents in [Mjolnir](https://github.com/BrokkAi/mjolnir/?ref=blog.slopcop.com). The price you pay is ~doubled task latency.\n\nAlternatively, you can drag DeepSeek Flash about 1/3 of the way to Opus in both intelligence and speed with the Oracle pattern, at the cost of doubling your price.\n\nAt SlopCop, we use cheaper subagents when we can't compromise on intelligence, and Oracle when that frontier intelligence is overkill but DeepSeek Flash isn't quite smart enough alone.", "url": "https://wpnews.pro/news/a-cheapskate-s-guide-to-coding-with-expensive-models", "canonical_source": "https://blog.slopcop.com/a-cheapskates-guide-to-coding-with-expensive-models/", "published_at": "2026-11-12 23:31:03+00:00", "updated_at": "2026-11-12 23:52:08.344063+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "large-language-models", "ai-products"], "entities": ["SlopCop", "Jonathan Ellis", "Mjolnir", "BrokkAi", "Opus", "Luna", "DeepSeek Flash", "Claude Code"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/a-cheapskate-s-guide-to-coding-with-expensive-models", "markdown": "https://wpnews.pro/news/a-cheapskate-s-guide-to-coding-with-expensive-models.md", "text": "https://wpnews.pro/news/a-cheapskate-s-guide-to-coding-with-expensive-models.txt", "jsonld": "https://wpnews.pro/news/a-cheapskate-s-guide-to-coding-with-expensive-models.jsonld"}}