--- title: Junie Local became 70% faster on Mac M5 | daily.dev description: Junie Local now runs coding tasks at about 70 tokens per second on Mac M5 Max, up from 40, with bursts to 107. The speedup comes from two changes: the Splash... canonical: https://daily.dev/posts/junie-local-became-70-faster-on-mac-m5-qh7khlxwd twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Junie Local became 70% faster on Mac M5 | daily.dev og:description: Junie Local now runs coding tasks at about 70 tokens per second on Mac M5 Max, up from 40, with bursts to 107. The speedup comes from two changes: the Splash... og:url: https://daily.dev/posts/junie-local-became-70-faster-on-mac-m5-qh7khlxwd og:image: https://api.daily.dev/og/posts/qH7KHLxwD.png og:image:alt: Junie Local became 70% faster on Mac M5 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. # Junie Local became 70% faster on Mac M5 **[JetBrains](https://daily.dev/sources/jetbrains)** · 4 min read · 1 upvotes · 0 comments ## Summary Junie Local now runs coding tasks at about 70 tokens per second on Mac M5 Max, up from 40, with bursts to 107. The speedup comes from two changes: the Splash inference engine, which replaces MLX-VLM, and a new DFlash2 draft model. Prefill improved from 850 to 1200 tokens per second, and a 100k context now needs about 24GB of RAM. Junie Local now also supports M4 Macs with 64GB RAM, though prefill is about 4 times slower there than on M5. Upgrade by running 'junie update' and then '/local' with ctrl+r. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How fast is Junie Local on an M5 Max now compared to before? Junie Local generates around 70 tokens per second on M5 Max for coding tasks, up from about 40 tokens per second previously, with bursts reaching 107 tokens per second. The gain comes from switching to the Splash inference engine and adding a new DFlash2 draft model fine-tuned on Qwen3.8-3.6-27B-blend, giving roughly a 2x improvement over the prior MTP plus MLX-VLM setup. _Developers tracking local AI coding agent speed can follow benchmarks like these on daily.dev._ ### Can I run Junie Local on an M4 Mac? Yes, Junie Local now supports M4 Macs and newer with at least 64GB of RAM. However, prefill is approximately 4 times slower on M4 than on M5, meaning loading the system prompt and context takes longer. On M5, prefill reached 1200 tokens per second, up from 850. _Anyone weighing Mac hardware for local AI coding workloads can track these requirements on daily.dev._ ### How much RAM does Junie Local need for a large context window? Junie Local needs about 24GB of RAM for a 100k token context after memory optimizations, down from prior consumption with the MLX-VLM setup. This reduction came alongside the switch to the Splash inference engine and the DFlash2 draft model, which together also improved generation speed roughly 2x. _Developers sizing hardware for local coding agents can keep tabs on changes like this via daily.dev._ ## Similar posts on daily.dev - [Junie Local: Smarter, Faster AI Coding](https://daily.dev/posts/junie-local-smarter-faster-ai-coding-dezqfgxuw) · JetBrains · 2 upvotes · 0 comments - [Running local models on an M4 with 24GB memory](https://daily.dev/posts/running-local-models-on-an-m4-with-24gb-memory-iezpeho6j) · Hacker News · 4 upvotes · 0 comments --- Tags: [#ai-inference](https://daily.dev/tags/ai-inference), [#local-ai](https://daily.dev/tags/local-ai), [#jetbrains](https://daily.dev/tags/jetbrains) [View this post on daily.dev](https://daily.dev/posts/junie-local-became-70-faster-on-mac-m5-qh7khlxwd) ```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":"Junie Local became 70% faster on Mac M5","url":"https://daily.dev/posts/junie-local-became-70-faster-on-mac-m5-qh7khlxwd","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/junie-local-became-70-faster-on-mac-m5-qh7khlxwd"},"datePublished":"2026-10-29T13:17:25.127Z","dateModified":"2026-10-29T13:18:22.250Z","description":"Junie Local now runs coding tasks at about 70 tokens per second on Mac M5 Max, up from 40, with bursts to 107. The speedup comes from two changes: the Splash...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/7f88d0a7080efdc194d94e3bf6744776?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/7f88d0a7080efdc194d94e3bf6744776?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"JetBrains","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":"JetBrains","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/53ecf0c676f34b3896ee109609d91efa","url":"https://daily.dev/sources/jetbrains"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/junie-local-became-70-faster-on-mac-m5-qh7khlxwd","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-inference,local-ai,jetbrains","timeRequired":"PT4M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"JetBrains","item":"https://daily.dev/sources/jetbrains"},{"@type":"ListItem","position":3,"name":"Junie Local became 70% faster on Mac M5"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/junie-local-became-70-faster-on-mac-m5-qh7khlxwd#faq","mainEntity":[{"@type":"Question","name":"How fast is Junie Local on an M5 Max now compared to before?","acceptedAnswer":{"@type":"Answer","text":"Junie Local generates around 70 tokens per second on M5 Max for coding tasks, up from about 40 tokens per second previously, with bursts reaching 107 tokens per second. The gain comes from switching to the Splash inference engine and adding a new DFlash2 draft model fine-tuned on Qwen3.8-3.6-27B-blend, giving roughly a 2x improvement over the prior MTP plus MLX-VLM setup. Developers tracking local AI coding agent speed can follow benchmarks like these on daily.dev."}},{"@type":"Question","name":"Can I run Junie Local on an M4 Mac?","acceptedAnswer":{"@type":"Answer","text":"Yes, Junie Local now supports M4 Macs and newer with at least 64GB of RAM. However, prefill is approximately 4 times slower on M4 than on M5, meaning loading the system prompt and context takes longer. On M5, prefill reached 1200 tokens per second, up from 850. Anyone weighing Mac hardware for local AI coding workloads can track these requirements on daily.dev."}},{"@type":"Question","name":"How much RAM does Junie Local need for a large context window?","acceptedAnswer":{"@type":"Answer","text":"Junie Local needs about 24GB of RAM for a 100k token context after memory optimizations, down from prior consumption with the MLX-VLM setup. This reduction came alongside the switch to the Splash inference engine and the DFlash2 draft model, which together also improved generation speed roughly 2x. Developers sizing hardware for local coding agents can keep tabs on changes like this via daily.dev."}}]} ```