--- title: I cancelled my Anthropic subscription for Qwen Token... description: A developer shares their experience switching from Anthropic Claude (Fable/Opus) to Qwen3.8-Max-Preview for their private software engineering workflow. They... canonical: https://daily.dev/posts/i-cancelled-my-anthropic-subscription-for-qwen-token-plan-because-of-qwen3-8-max-preview-ln2gnx7po twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: I cancelled my Anthropic subscription for Qwen Token Plan because of Qwen3.8-Max-Preview | daily.dev og:description: A developer shares their experience switching from Anthropic Claude (Fable/Opus) to Qwen3.8-Max-Preview for their private software engineering workflow. They... og:url: https://daily.dev/posts/i-cancelled-my-anthropic-subscription-for-qwen-token-plan-because-of-qwen3-8-max-preview-ln2gnx7po og:image: https://api.daily.dev/og/posts/LN2GNx7po.png og:image:alt: I cancelled my Anthropic subscription for Qwen Token Plan because of Qwen3.8-Max-Preview 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. # I cancelled my Anthropic subscription for Qwen Token Plan because of Qwen3.8-Max-Preview **[nerdalytics](https://daily.dev/sources/jcwojcedqzpspqbvk73nc)** · [@nerdalytics](https://daily.dev/nerdalytics) · 5 min read · 4 upvotes · 19 comments ## Summary A developer shares their experience switching from Anthropic Claude (Fable/Opus) to Qwen3.8-Max-Preview for their private software engineering workflow. They detail specific failures — a reasoning rigidity bug where Qwen anchored on a wrong hypothesis across sessions, and context misreading on deployment targets — while highlighting strengths like reliable plan execution, no weekly usage limits, screenshot understanding, and autonomous tool selection. They cancelled their Anthropic subscription in favor of Qwen Token Plan at the same price, while keeping GLM 5.2 for security auditing and noting Claude still edges out Qwen on certain hard reasoning problems. ## Content I've been using Qwen models since September last year. On Qwen Studio chatting mostly about software engineering, check in to stay on track, do deep research, and take the PDF output, and feed it into Qwen Code for brainstorming or execution. Over ten months that's a few hundred conversations I guess. When Qwen3.8-Max-Preview dropped, I used it daily for a few days before writing this. It's the first Qwen model that replaced Fable and Opus in my actual private workflow rather than sitting next to them. ## What I'm building A git code review solution (more on that soon). Qwen3.8-Max-Preview handles web search, ideations, or just discuss anything. Qwen Code handles execution. I run obra@superpowers skills for structured workflows: the model picks the right skill, spawns agents, digs into problems, proposes fixes. The fixes work. I run them past Fable, Opus, GLM 5.2, and myself as judges. Rarely does anyone find an issue after task reported "finished". Fable never failed at plan execution either. But Anthropic's subscription has weekly usage limits. Qwen Token Plan doesn't. That changed my week. ## Two things it got wrong I want to be specific because vague praise helps nobody. First: I had a bring-to-production plan designed by Fable 5. Qwen3.8-Max-Preview kept treating my MacBook environment details as an end-user setup. I had to interrupt several times and steer it toward treating my local setup as the deployment target. I don't blame the model. The plan Fable wrote never defined what "bring to production" meant relative to the machine running the task. Qwen got a spec with a hole in it and filled the hole with a reasonable default—which I disagreed with, as my dev machine is not the production environment. Second: a file:// protocol bug. I described that providing a path via file:// doesn't work, and stripping "file:" fixes it. Qwen latched onto the triple slash in "///path" and concluded the software was producing a malformed path. It carried that conclusion across multiple sessions. Opus, on first attempt, recognized that file:/// and /// differ only by the "file:" prefix and that this is user error, not a software bug. Opus proposed accepting the illformed path and educating users. Qwen only arrived at the same fix after I explicitly told it I had stripped "file:" from the URL instead of "file://". At that point it proposed the same change Opus had, minus the user education part. That second one is a reasoning rigidity failure. It anchored on a hypothesis and didn't question the premise across sessions. Opus saw it immediately. ## Where other models still win For security auditing, I use GLM 5.2. It consistently finds more issues than Qwen. Different training emphasis, different strengths. For deep research I use Qwen3.7 models, as Qwen3.8-Max-Preview doesn't have deep research capabilities on Qwen Studio. ## A note on Qwen Code and memory Qwen Code no longer looks in $HOME for AGENTS.md or similar grounding files. I noticed it wasn't picking up mine. What I didn't notice was Qwen failing to ground itself. The opposite, actually. In the thinking output I could watch Qwen remind itself to ground its statements and provide evidence before assuming blindly. That behavior is defined in my `$HOME/AGENTS.md`. Qwen internalized it from earlier 3.7-max sessions and kept following it from memory, file or no file. I assume the real change is that Qwen Code stopped looking for CLAUDE.md-style files in $HOME, but I'm not certain. Either way, the memory system carried the instructions forward. That impressed me more than the file lookup bug annoyed me. ## Positive surprise Screenshots. Qwen3.8-Max-Preview can see them. Debugging UI quirks used to mean describing pixel positions in text. Now I paste a screenshot and say "this button is wrong." Hours saved, repeatedly. In Qwen Studio, it one-shotted an egg cooking timer, a global satellite tracker pulling real-time data, a dino tamagotchi, and a handful of other small apps. See: [https://chat.qwen.ai/s/deploy/t_3a6f9977-50d0-4209-b6a9-f8105e7d9163](https://chat.qwen.ai/s/deploy/t_3a6f9977-50d0-4209-b6a9-f8105e7d9163) [https://chat.qwen.ai/s/deploy/t_46d8879d-3b74-4370-8164-33dd4cbf57f5](https://chat.qwen.ai/s/deploy/t_46d8879d-3b74-4370-8164-33dd4cbf57f5) [https://chat.qwen.ai/s/deploy/t_74e89599-1efc-4a5d-a85b-5b162187815f](https://chat.qwen.ai/s/deploy/t_74e89599-1efc-4a5d-a85b-5b162187815f) My deep research workflow, where Qwen produces long PDF reports that I feed into Qwen Code as brainstorming input, currently runs on Qwen3.7-Plus in Studio. Qwen3.8-Max-Preview doesn't have deep research in Qwen Studio yet. I'm waiting for full release to see what 3.8 does with it. If the reasoning quality carries over, those PDFs are going to get even better. Throughput is still settling. On release day I saw 80+ tokens per second. A few days later it's an unstable mix between 30 and 50 tps. I assume the preview isn't deployed on all available compute yet. Curious what the final numbers look like. ## The money part I cancelled my private Anthropic subscription and signed up for Qwen Token Plan Team Max. Same price. To be clear: I still have access to Fable and Anthropic models at work. This isn't a case of losing access and making do with what's left. I had both, compared them on the same projects, and chose Qwen for my private work. Added bonus: the Token Plan includes access to other Chinese open-weight models, GLM 5.2 among them. So I didn't lose my security auditing tool. ## What I'd tell someone on the fence You will correct Qwen3.8-Max-Preview about as often as you correct Fable or Opus. It fixates on wrong hypotheses sometimes. It misreads context when the spec is ambiguous. But it executes plans reliably, picks the right tools autonomously, produces fixes that pass review, and doesn't run out on a weekly timer. For my workflow, that's enough. It approaches problems differently from the Anthropic models, yet familiar enough that I didn't have to relearn how to work with it. If your work is security-heavy, keep GLM 5.2. If you need Fable's specific creative insight on a hard problem, Fable is still there. But for day-to-day software engineering, research, planning, debugging, and code review, Qwen3.8-Max-Preview took the seat and hasn't given it back. ## Community discussion Top comments from developers on daily.dev. **@geniusatwork** · 2 upvotes > Qwen is good. I can't effort the Max team token plan, but the Individual Pro plan I can. > > > Qwen code is good. The build-in pr-review skill is really good. But `qwen serve` is my favorite. Accessing all my Qwen sessions from a browser is great. > > 3.8-Max is fantastic. Its plan is usually lightweight. Then it goes and implements the plan. Compared to Fable, Qwen-3.8 doesn't enter review&fix loops so often. Likewise with Superpowers skills. Brainstorming, design, plan, execute, and finish. Fable reports it found issues during review, sent agent to fix, then Fable happily reports all green and... **@agustinbarrientos** · 2 upvotes > I'd change one assumption and hand the same bug right back to it. If it drops the old theory, great. If it doubles down, I'm not wasting another round arguing with it. **@randy** · 0 upvotes > Enjoy having your IP (potentially) stolen... --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#anthropic](https://daily.dev/tags/anthropic), [#qwen](https://daily.dev/tags/qwen), [#glm](https://daily.dev/tags/glm) [View this post on daily.dev](https://daily.dev/posts/i-cancelled-my-anthropic-subscription-for-qwen-token-plan-because-of-qwen3-8-max-preview-ln2gnx7po) ```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":"DiscussionForumPosting","mainEntityOfPage":"https://daily.dev/posts/i-cancelled-my-anthropic-subscription-for-qwen-token-plan-because-of-qwen3-8-max-preview-ln2gnx7po","headline":"I cancelled my Anthropic subscription for Qwen Token Plan because of Qwen3.8-Max-Preview","text":"A developer shares their experience switching from Anthropic Claude (Fable/Opus) to Qwen3.8-Max-Preview for their private software engineering workflow. They detail specific failures — a reasoning rigidity bug where Qwen anchored on a wrong hypothesis across sessions, and context misreading on deployment targets — while highlighting strengths like reliable plan execution, no weekly usage limits, screenshot understanding, and autonomous tool selection. They cancelled their Anthropic subscription in favor of Qwen Token Plan at the same price, while keeping GLM 5.2 for security auditing and noting Claude still edges out Qwen on certain hard reasoning problems.","url":"https://daily.dev/posts/i-cancelled-my-anthropic-subscription-for-qwen-token-plan-because-of-qwen3-8-max-preview-ln2gnx7po","datePublished":"2026-07-23T07:33:00.122Z","dateModified":"2026-09-13T21:14:05.127Z","author":{"@type":"Person","name":"nerdalytics","url":"https://daily.dev/nerdalytics","image":"https://avatars.githubusercontent.com/u/97166791?v=4","description":"Husband. Cat and dog dad. Nerd.","interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"EndorseAction"},"userInteractionCount":4610}},"image":"https://media.daily.dev/image/upload/s--fewczlRJ--/f_auto/v1784790338/posts/LN2GNx7po?_a=BAMAMicg0","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":4},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":19}],"comment":[{"@type":"Comment","text":"Qwen is good. I can’t effort the Max team token plan, but the Individual Pro plan I can.\nQwen code is good. The build-in pr-review skill is really good. But qwen serve is my favorite. Accessing all my Qwen sessions from a browser is great.\n3.8-Max is fantastic. Its plan is usually lightweight. Then it goes and implements the plan. Compared to Fable, Qwen-3.8 doesn’t enter review&fix loops so often. Likewise with Superpowers skills. Brainstorming, design, plan, execute, and finish. Fable reports it found issues during review, sent agent to fix, then Fable happily reports all green and presents a summary of all divergences from the plan. Often with strange assumptions. Qwen doesn’t do that often. It feels like Qwen sticks to the goal more closely. Much less surprises about how the software changed with Qwen. Fable has these moments of brilliance, but those are overshadowed with these many moments of WTF decisions that Fable did. Fable is more useful for UI work. Because of all the decisions Fable makes during implementation, UI looks a better. For non UI work Fable gets under my skin with all the reading between lines. I didn’t need backwards compatibility. I didn’t need Raft protocol derivative. I didn’t need a self rolled Bash test suite for this 30 lines Bash script.\nSome people like that, I certainly don’t. Qwen is more my type of behavior. Happy Qwen exists.","datePublished":"2026-08-22T06:45:31.227Z","url":"https://daily.dev/posts/LN2GNx7po#c-f09lnv7lA","author":{"@type":"Person","name":"Sam Dolin","url":"https://daily.dev/geniusatwork","image":"https://media.daily.dev/image/upload/s--nUhoZxKm--/f_auto/v1787434895/avatars/avatar_75QyPrL3BZgArSo6aFWuZ?_a=BAMAMicg0"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2}},{"@type":"Comment","text":"I’d change one assumption and hand the same bug right back to it. If it drops the old theory, great. If it doubles down, I’m not wasting another round arguing with it.","datePublished":"2026-09-13T03:29:41.490Z","url":"https://daily.dev/posts/LN2GNx7po#c-6Oj71e3W6","author":{"@type":"Person","name":"Agustin Barrientos","url":"https://daily.dev/agustinbarrientos","image":"https://media.daily.dev/image/upload/s--5ayxQnqn--/f_auto/v1788281802/avatars/avatar_wQYYVe5Tbj0NJ7C7qPoa8?_a=BAMAMicg0"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2}},{"@type":"Comment","text":"Enjoy having your IP (potentially) stolen…","datePublished":"2026-07-23T13:38:01.587Z","url":"https://daily.dev/posts/LN2GNx7po#c-LWv7ulRl9","author":{"@type":"Person","name":"Randy","url":"https://daily.dev/randy","image":"https://media.daily.dev/image/upload/s--UjV4-KkB--/f_auto/v1708097210/avatars/avatar_HXYbbGcBO38Rfv7RrCBdA"}}],"isPartOf":{"@type":"WebPage","url":"https://daily.dev/sources/jcwojcedqzpspqbvk73nc","name":"nerdalytics"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"nerdalytics","item":"https://daily.dev/sources/jcwojcedqzpspqbvk73nc"},{"@type":"ListItem","position":3,"name":"I cancelled my Anthropic subscription for Qwen Token Plan because of Qwen3.8-Max-Preview"}]} ```