--- title: Devin CLI adds Fusion mode: mix planning and execution... description: Cognition has launched Fusion mode in Devin CLI, a multi-model configuration that pairs a stronger model for planning with a cheaper model for execution.... canonical: https://daily.dev/posts/devin-cli-adds-fusion-mode-mix-planning-and-execution-models-to-cut-costs-lq8fdh4yz twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Devin CLI adds Fusion mode: mix planning and execution models to cut costs | daily.dev og:description: Cognition has launched Fusion mode in Devin CLI, a multi-model configuration that pairs a stronger model for planning with a cheaper model for execution.... og:url: https://daily.dev/posts/devin-cli-adds-fusion-mode-mix-planning-and-execution-models-to-cut-costs-lq8fdh4yz og:image: https://api.daily.dev/og/posts/lq8fDH4YZ.png og:image:alt: Devin CLI adds Fusion mode: mix planning and execution models to cut costs 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. # Devin CLI adds Fusion mode: mix planning and execution models to cut costs **[Collections](https://daily.dev/sources/collections)** · 2 min read · 1 upvotes · 1 comments ## Summary Cognition has launched Fusion mode in Devin CLI, a multi-model configuration that pairs a stronger model for planning with a cheaper model for execution. Cognition claims this setup runs 39% cheaper across coding benchmarks while still leveraging frontier models like Fable and Astra. Artificial Analysis independently benchmarked Devin Fusion on release day, marking the first third-party evaluation of a multi-model coding agent of this kind. The core idea is to reserve expensive reasoning for planning steps while offloading execution to lower-cost models, reducing overall task costs. ## Content Cognition just released Fusion in the Devin CLI, a setup that lets you assign one model for planning and a separate, cheaper model for execution. The pitch is straightforward: you get close to the same benchmark scores at meaningfully lower cost. The numbers from Artificial Analysis's independent benchmark tell the story pretty clearly: - Fable 5.1 High as lead + SWE-2 Medium as sidekick in Fusion: **61.7 score at $7.90/run** - Fable 5.1 solo in Claude Code: **62.2 score at $12.36/run** That's less than one point difference on the coding agent index, but 36% cheaper per run. If you want to go even cheaper, swapping Astra as the lead drops the cost further: - Astra + SWE-2 in Fusion: **58.9 score at $4.54/run** - Astra solo in Codex: **61.6 score at $7.47/run** That's 39% cheaper, though you give up about 2.7 points on the benchmark. One user reported spending five hours running Fable 5.1 High across four repos and only hitting 15% of their weekly Devin Max quota. Whether that holds up at scale is worth watching, but the efficiency angle seems real. To get started: install the Devin CLI or desktop app, select Fusion, set Fable 5.1 High as the lead model and SWE-2 Medium as the sidekick, and turn off fast mode. ## Questions this post answers ### What is Fusion mode in Devin CLI? Fusion mode is a multi-model configuration in Devin CLI that pairs one model for planning with a cheaper model for execution. Cognition states it runs 39% cheaper across coding benchmarks while still using frontier models like Fable and Astra for the reasoning-heavy planning steps, offloading execution to a more cost-effective model. _Developers weighing coding agent costs against capability can track releases like this one on daily.dev._ ### How much cheaper is Devin Fusion compared to using a single frontier model for coding tasks? Cognition reports Devin Fusion runs 39% cheaper across coding benchmarks compared to using a single frontier model for every step. The savings come from reserving expensive planning models like Fable and Astra for reasoning while routing execution work to a lower-cost model. _Teams comparing coding agent pricing models can follow updates like this through daily.dev._ ### Has any third party independently verified Devin Fusion's benchmark claims? Yes, Artificial Analysis independently benchmarked Devin Fusion on its release day, marking the first time a multi-model coding agent has been evaluated by a third party in this way. This provides outside validation alongside Cognition's own reported cost savings figures. _Anyone deciding between AI coding agents can follow independent benchmark coverage via daily.dev._ ## Community take How the wider developer community reacted, aggregated from 1 discussion and 30 comments across x (as of 2026-09-15). **TL;DR:** Reactions to Fusion mode are largely enthusiastic about cost savings, especially since one of the execution models is currently free, though several people ask pointed questions about how work is actually split between lead and sidekick and whether quality holds up. **Sentiment:** 40% positive · 45% mixed · 15% skeptical **The case for** - Several users report barely using their usage budget after hours of work across multiple repos. - The idea of routing cheap models for volume work and reserving the expensive model for judgment calls is seen as the key insight. - One combo is currently free to use, making the cost savings even more attractive to try. **The pushback** - Some ask whether the benchmark numbers are inflated by SWE-2 being offered for free at the moment, which wouldn't reflect real-world cost. - Questions remain about which model actually handles file reading and repo context loading, since that affects true cost savings. - One commenter flags that the benchmark screenshot uses the recommended 'High' effort setting rather than the actual tested configuration, calling the comparison into question. - Uncertainty about whether output quality from the fusion pairing matches using the flagship model alone, especially on longer tasks. **By community** - x (positive): Casual, mostly upbeat replies praising the cost efficiency and low usage impact, mixed with practical questions about how the lead/sidekick split actually works and skepticism about the free-pricing skewing benchmarks. **Hottest debate:** Whether the benchmark cost/quality numbers are meaningfully undercut by one of the execution models currently being offered for free. **Open questions** - Which model actually performs file reading and repo-loading work in a fusion setup, and does that cost get hidden? - How much can the fusion combo be used under lower-tier subscription plans before hitting limits? - Does fusion output quality match using the flagship model solo, particularly on longer or more complex tasks? **Highlights** > @kloss_xyz @DevinAI Fusion looks excellent, and I agree that Fable with SWE-2 is probably the pairing most people should try first. The only part I would correct is the effort level attached to the benchmark scores. The High setting in your screenshot is your recommended setup, not the configuration > — [GodsBoy7777 on x](https://x.com/GodsBoy7777/status/2099804798019080224) > @kloss_xyz @DevinAI I do not think they let fable do the main job, swe is likely the orchestrator and reach out to fable when necessary. Because I am using the smallest plan it feels more obvious, the usage was not moving when it was reading stuff and checking stuff, then a big drop. > — [Eloitay on x](https://x.com/Eloitay/status/2099835028440317967) > @kloss_xyz @DevinAI Are you taking under concideration the fact that SWE-2 is being offered for free atm? > — [Sami\_Sytela on x](https://x.com/Sami_Sytela/status/2099759579609420098) > @kloss_xyz @DevinAI Which side reads the four repos: does SWE-2 do the file reading, or does Fable still pay to load each one before it hands off? > — [use\_kibble on x](https://x.com/use_kibble/status/2099743690852118693) > @kloss_xyz @DevinAI lead plans, sidekick executes is the whole trick. flagship for judgment, cheap for volume. running flagship for everything is just donating margin > — [foggybuild on x](https://x.com/foggybuild/status/2099837530078319101) **Source threads** - [x](https://x.com/kloss_xyz/status/2099714012867178531) · 0 points · 30 comments ## Community discussion Top comments from developers on daily.dev. **@fayzuralmas** · 1 upvotes > **The smartest agents won’t be the ones that always use the strongest model they’ll be the ones that know when not to.** ## Similar posts on daily.dev - [Fable 5 vs GPT-5.6 Sol: I spent $2,000 and 2 billion tokens to find out who wins — Expo blog](https://daily.dev/posts/fable-5-vs-gpt-5-6-sol-i-spent-2-000-and-2-billion-tokens-to-find-out-who-wins-expo-blog-hfefgd2ti) · Expo Blog · 0 upvotes · 0 comments --- Tags: [#ai](https://daily.dev/tags/ai), [#ai-agents](https://daily.dev/tags/ai-agents), [#devin](https://daily.dev/tags/devin) [View this post on daily.dev](https://daily.dev/posts/devin-cli-adds-fusion-mode-mix-planning-and-execution-models-to-cut-costs-lq8fdh4yz) ```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":"Devin CLI adds Fusion mode: mix planning and execution models to cut costs","url":"https://daily.dev/posts/devin-cli-adds-fusion-mode-mix-planning-and-execution-models-to-cut-costs-lq8fdh4yz","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/devin-cli-adds-fusion-mode-mix-planning-and-execution-models-to-cut-costs-lq8fdh4yz"},"datePublished":"2026-09-11T23:27:46.555Z","dateModified":"2026-09-15T12:32:56.334Z","description":"Cognition has launched Fusion mode in Devin CLI, a multi-model configuration that pairs a stronger model for planning with a cheaper model for execution....","isAccessibleForFree":true,"articleSection":"Collections","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":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":1,"discussionUrl":"https://daily.dev/posts/devin-cli-adds-fusion-mode-mix-planning-and-execution-models-to-cut-costs-lq8fdh4yz","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":1}],"keywords":"ai,ai-agents,devin","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"Devin CLI adds Fusion mode: mix planning and execution models to cut costs"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/devin-cli-adds-fusion-mode-mix-planning-and-execution-models-to-cut-costs-lq8fdh4yz","comment":[{"@type":"Comment","text":"The smartest agents won’t be the ones that always use the strongest model they’ll be the ones that know when not to.","datePublished":"2026-09-12T09:02:27.258Z","dateModified":"2026-09-12T09:02:50.377Z","url":"https://daily.dev/posts/lq8fDH4YZ#c-SzSOQIXp4","author":{"@type":"Person","name":"fayzur almas","url":"https://daily.dev/fayzuralmas","image":"https://media.daily.dev/image/upload/s--S83QkUhg--/f_auto/v1778866221/avatars/avatar_0cQh2iYcRC1iyoBZLhPy8?_a=BAMAMiWQ0"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1}}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/devin-cli-adds-fusion-mode-mix-planning-and-execution-models-to-cut-costs-lq8fdh4yz#faq","mainEntity":[{"@type":"Question","name":"What is Fusion mode in Devin CLI?","acceptedAnswer":{"@type":"Answer","text":"Fusion mode is a multi-model configuration in Devin CLI that pairs one model for planning with a cheaper model for execution. Cognition states it runs 39% cheaper across coding benchmarks while still using frontier models like Fable and Astra for the reasoning-heavy planning steps, offloading execution to a more cost-effective model. Developers weighing coding agent costs against capability can track releases like this one on daily.dev."}},{"@type":"Question","name":"How much cheaper is Devin Fusion compared to using a single frontier model for coding tasks?","acceptedAnswer":{"@type":"Answer","text":"Cognition reports Devin Fusion runs 39% cheaper across coding benchmarks compared to using a single frontier model for every step. The savings come from reserving expensive planning models like Fable and Astra for reasoning while routing execution work to a lower-cost model. Teams comparing coding agent pricing models can follow updates like this through daily.dev."}},{"@type":"Question","name":"Has any third party independently verified Devin Fusion's benchmark claims?","acceptedAnswer":{"@type":"Answer","text":"Yes, Artificial Analysis independently benchmarked Devin Fusion on its release day, marking the first time a multi-model coding agent has been evaluated by a third party in this way. This provides outside validation alongside Cognition's own reported cost savings figures. Anyone deciding between AI coding agents can follow independent benchmark coverage via daily.dev."}}]} ```