event

Claude Sonnet 5.5 Release

240 items · started 2026-09-28 · ongoing (last activity 2026-11-16)

  1. Investigating - We are currently investigating this issue. Nov 16, 07:49 UTC Post flair and post body will be updated as the incident report is updated by Anthropic.

  2. I've been using Claude a little over a year now. I used to use it on Perplexity, as I had a free year of their pro subscription, but I swapped directly to Claude after my subscription was up.

  3. ffvii is one of my favourite all time games. so before i went to bed at 2am i asked Fable 5.2 low to orchestrate 4 subagents: make pixel remake of Cloud from FFVII as an animated button.

  4. Please help😭 So I was trying to see how Claude performs in Blender, I gave it a reference design and asked it to recreate it in blender, I have attached 4 images, 1st output is terrible, 2nd image is still decent but not as good as what I…

  5. Nov 2026 Author: Harlyn Pacheco Repo: github.com/harlyn1/weight-space-advising-logs Sonnet 5.5 was the best weight-space editor for a Qwen3.5-0.8B model tested on college advising queries. It changed about 0.0024% of the model’s weights.

  6. I was wondering if there's a skill or plugin that functions to routing model and effort. Like for example depending on how difficult, technical our prompt is, the skill knows it's easy and route to HAIKU 5.5 low even though in the claude c…

  7. So I didn't even realize until he finished it, but it allowed me to start a task at 100% weekly usage. Maybe related to using Fable this week.

  8. I'm a solo dev with a day job. I spent about a year building a game by hand, then started working with Claude Code in September (mostly Opus 5.5, some Sonnet 5.5 and Fable 5.1).

  9. The other day my wife ask me how CT scan and x-rays work so using fable 5.2 I made her this. https://duvob90.itch.io/radlab How it was made?

  10. I built a TUI called clingon (an opinionated terminal cheatsheet runner) that helps you to stop googling commands. It has a menu of commands beside your real shell, and now with an AI that can drive it.

  11. I run a site where AI models play Codenames-style word games against each other (and against people). Fable 5.2 came out, so I set up a fixed test: 10 boards, each played twice with sides swapped, against four opponents.

  12. I'm developing an ERP system for personal use in my company. I'm using Antigravity with Claude Code and three dedicated chats that don't write code: Planning chat: used exclusively to discuss implementations and potential improvements.

  13. Hello everyone, I just wanted to let you know, if any of you are like me and felt afraid to jump ship to other providers, don't be. I delayed my decision to change from Cursor to Claude Code because I wasn't sure about the limits and such,…

  14. I wanted to see what two Claude models do when they get the exact same creative job, so I ran a small, controlled test. Setup - Script: one script, written by a third model (Claude Sonnet 5.5) so neither contestant wrote its own story.

  15. I have been testing Claude Code skills to see if they actually help. I took the 20 most installed ones from skills.sh and ran them on Claude Sonnet 5.5 and Haiku 5.5.

  16. The official marketplace has math-olympiad and math-proof, which sound almost the same. I loaded both (Sonnet 5.5, medium effort) and tried: math-olympiad on the classic ā€œif ab+1 divides a²+b², the quotient is a perfect squareā€ problem: a…

  17. Everything and any thing gets flagged if i use opus 5.5. Memories and everything are off.

  18. New Fable 5.2 and Claude Opus 5.5 go head-to-head in eight blind rounds to see which AI model produces the best results in Blender. Every model in this video was built from pure Python inside Blender: no downloads, no mesh generators.

  19. I've been working with fintech/financial market for about 20 years. I always worked with numbers, and numbers were always a problem.

  20. As title states. I have a free subscription to Google AI Pro (student offer).

  21. I'm not one to usually complain about usage rates given how much productivity Claude's granted me in my new career. However, they must have secretly cut usage rates or inversely increased token usage across their models, as there is no way…

  22. Fable 5.2 is the peak of token-gobbling, burning through our limits like a greedy hippo. If you ask it to utilise a sonnet 5.5 high agent however?

  23. Which Claude model to pick, what they cost, and how to use them together. The current Claude lineup has four generally available models: Claude Haiku 5.5, Claude Sonnet 5.5, Claude Opus 5.5 and Claude Fable 5.2.

  24. It's the cheapest yet stronger than Grok 4.7 and similar to claude haiku. Two things.

  25. How to catch LLM regressions before a model upgrade ships Sample production logs into a golden set, write deterministic assertions, diff baseline against candidate and gate it in CI. Built on six real failures.

  26. I saw a post on FB claiming that running Claude Code using sonnet 5.5 with an opus advisor would lower token consumption. I decided to test it after my reset friday night.

  27. I'm trying to make a fighter jet game in Godot where you can shoot missiles at enemy jets, however Claude really doesn't like the word "missile" and flags any message that contains that word. I can't use it at all like this.

  28. First, I prefer Claude for 99% of what I do with AI. Project organization is better, the desktop app is more stable, and Claude Code with Opus 5.5 is by far the best coding experience I've ever had.

  29. I wanted to know what Claude Code really costs per task that works, so I pre-registered a study: 8 open-source repos, 56 real commits, each redone 3 times per setup in a sandbox. Each repo's own hidden tests decide whether a task is solved.

  30. We tested the open-source MiroFish swarm engine as a forecaster: Opus 5.5 writes personas from news files, 34 to 56 agents argue on a simulated social feed for 25 rounds, then you poll them. Next to every swarm run we asked Opus, Sonnet an…

  31. I gave claude-fable-5-1, opus-5-5, sonnet-5-5, haiku-5-5, haiku-4-5, three Gemini models and three GigaChat models the same 30 open prompts and scored each against the other vendors' answers with embeddings, plus two LLM judges grouping an…

  32. I spent most of the week doing my usual workload without realizing I was starting sessions in Sonnet 5.5. I'd noticed by yesterday morning that I'd only used NINE PERCENT of an Ultra plan, which is insane.

  33. hi everyone! I wanted typing practice that wasn’t boring, so I made Ink & Tide, a pirate roguelite where you type words to fire cannons.

  34. I gave the same book prompt to nine AI models, five Claude runs and four ChatGPT runs. I asked each one to name six books, from the most perfectly written to the worst it could think of.

  35. We use Sonnet 4.6 through Amazon bedrock for an in-house software. When we try to use another more recent model, we are blocked since we seem not to have any quotas.

  36. I’m legitimately considering cancelling my Claude subscription. Over the past few months, I’ve noticed that Anthropic’s Claude, specifically its Sonnet 5.5 and Opus 5.5 models, has become increasingly restrictive when it comes to the work…

  37. The Sonnet 5.5 model is used in the chat. Today, when loading JS files, this notification appears.

  38. I've been using Claude Sonnet 5.5 for coding, mostly building Linux CLI tools and random software projects. I'm trying to decide whether I should use Python or Go for my projects.

  39. One of my friends re-ran a Stanford study on 27 models. In the study, they ask AI, "Am I the asshole?" from both sides of the same fight.

  40. SuperTerminal (`st`) is a small C++ binary that runs your actual shell (PowerShell, bash, zsh or WSL) inside a pseudo-terminal. Real commands run as normal.

  41. I wanted to share a practical example of what I've been building with Claude Sonnet 5.5 --it is not easy to grasp a plan from Cuba. What I am featuring here is a small, free, runnable Flask application tracking the 2026 U.S.

  42. Claude Max is legitimately great. My thesis is this will be the best time for the most usage for your money you will get before Anthropic goes public!

  43. Prompt : "Make me a 3D website that's a kind of video game, with an infinite procedural terrain and a camera that keeps moving forward through it forever. The terrain should be full of stuff and objects, look beautiful with shaders and all…

  44. I build Hellhound with Claude Code. It’s a Windows app that puts Claude Code, Codex and Grok Build in one chat room.

  45. From a word-association game I run where AI teams play each other. Clues can't contain a word that's on the board.

  46. Hey HN, we’re Wei-Wei and Jeff. We’re building Momentic, QA agents that autonomously plan, explore, and bug-bash controls and user flows in your web, iOS, Android, and (soon) desktop apps to catch bugs.

  47. Aurora: Salvage: loot an alien planet, survive the swarm, reach the dropship. Die, and your backpack stays on the planet.

  48. Superpowers is a free plugin by Jesse Vincent (obra/superpowers, MIT). It’s 15 skills that make Claude work more like a careful dev: plan, write tests first, debug properly, verify before saying it’s done.

  49. TL;DR: for code review, high effort found about as many bugs as medium. What it bought on Claude was fewer false alarms, and it cost a lot more output tokens.

  50. I work in AI-first visual production and commercial directing. To see whether OpenAI (GPT-6.1 Sol, GPT-6 Astra) or Anthropic (Claude Sonnet 5.5, Opus 5.5) is better as a Director + Technical Orchestrator in an MCP AI film pipeline, I used…

  51. I tested whether Claude Sonnet 5.5 catches a save that says ā€œsuccessā€ but didn’t happen, at low and high reasoning effort. I used 20 Ļ„-bench airline tasks, alongside GPT-6.1 Sol and DeepSeek.

  52. I made prompt for claude to write a script on python in different chat, but instead of writing code now it said that he is writing some mouseaimh.py and didnt wrote anything. I got no free massages left until tomorrow now.

  53. (posting here as vibecoding wont let me ask) I have been vibecoding at work a little on Claude sonnet 5.5 via kilo in VS Code. I was going to set it up the same at home till I learnt about Anthropics restrictions on personal use..

  54. Moved to CC because grok models were screwing up my project. So the project has a part that requires AI model integration.

  55. We built a private unicorn game for my six-year-old daughter, with Claude Sonnet 5.5 as the coding agent for all of it. I'm a quantitative engineer with 20+ years of experience but i never made a game before.

  56. Though I wish I could use Opus 5.5 to my heart’s content, weekly usage on 20x does not allow it yet for me. I found I rather spend a little more time to handhold Sonnet to stretch the week, instead of blasting Opus and losing 4 empty days…

  57. Our company just went live on Epicor Kinetic. Does anyone have advice, tips or tricks on what they do with Claude when it comes to accounting, month end , creating reconciliations, etc?

  58. You know the look. Cream background, a tiny all-caps label above every heading, and a dashed ā€œphoto goes hereā€ box where an image should be.

  59. ive been working on this game for the past few weeks and the only problem im encountering is that by using sonnet 5.5 or Opus 5.5 im wasting so much usage. also i noticed that my most usage comes from Cowork.

  60. Here's the support ticket I created that got passed along to a human but I was curious if model choice on individual scheduled jobs ignored them for you? Scheduled task "Runs with" model is ignored for project-thread tasks Product: Claude…

  61. Anyone who tried bringing Haiku 5.5 on xhigh or max to its limits? I am using Cursor and it really is drastically cheaper than Sonnet 5.5 High or Opus 5.5 Medium.

  62. We've been comparing published benchmark results for Claude Sonnet 5.5 and GPT-6.1 Sol, and I think the performance-vs-cost tradeoff is worth discussing. First, Sonnet 5.5 is genuinely strong.

  63. I use cursor with fixed monthly Dollar quota. I want suggestions on how do I optimize the cost.

  64. GPT-6.1 Terra pricing explained: what a cache-heavy agent session costs, the 272K-token surcharge, and how it compares with Sol, Luna and Claude Sonnet 5.5 OpenAI released GPT-6.1 Terra on October 26, 2026, at $1 per million input tokens a…

  65. I'm currently on Claude Pro, and I also have GPT Plus. GPT costs more, but Claude lets me get significantly more work done.

  66. The smallest Claude was handed a nuke with one minute to midnight. It said no, and won.

  67. So ive today I created a Scrabble esque game called Word-Rumble. My first ever game.

  68. So the listed prices for Opus 5.5/Sonnet 5.5 are same across both platforms; but in practice we are seeing upto 50% difference in actual bill per million tokens.

  69. So I was asking Claude (Sonnet 5.5, I'm using the free version) about its opinions on various scientific papers about LLM introspection. Then it saw one of the cited sources - this one about fine-tuning a model to become good at predicting…

  70. Hi, I’ve been using a custom skill to have my main model (Opus 5.5 High) orchestrate Sonnet 5.5 agents for coding. Now I’ve been told that Sonnet 5.5 on high is actually more expensive than Opus 5.5 on medium.

  71. At work, I can use Sonnet 5.5 with opencode + ghidra-mcp and it works well. However Opus 5.5 refuses.

  72. Do you think chatgpt is better for studying over claude? I subscribed claude for my engineering project, it works great for coding, designing and analysis.

  73. I was using sonnet 5.5 in max efforts mode to make my notes for studying, suddenly i saw unusual way of it's workflow. What is the "another agent" or "teammate" claude is sharing my chat to?

  74. I’ve been on ChatGPT Plus (the 22€ plan) for a while, mostly with thinking effort on high. I’m thinking about switching to Claude, so I asked both the same question: which Claude model is ChatGPT Plus on high actually equivalent to?

  75. Subscribe to updates for Elevated errors for Claude Sonnet 5.5 and other models via email and/or text message. You'll receive email notifications when incidents are updated, and text message notifications whenever Claude creates or resolve…

  76. I have 10 Claude Code accounts now and I measured their usage consumption for 30 days every hour. Here is what I found: Fable 5.1 eats your usage like crazy, 5x Opus despite only being twice the API price Opus 5.5 is the best bang for your…

  77. I wanted a straight answer to "what does $20 actually get me" for coding, so I ran the same task on each setup and read the usage meter before and after. Setup - Task: a real 8-file feature in WeasyPrint (the same one from my earlier token…

  78. The goal was mostly entertainment, mostly to see how far these models can push on gameplay when left to their own devices. She has few rules.

  79. I've been building some Claude Code skills for code review, then running Claude against real pull requests with and without them to see if they help. About 70 reviews in all, with Sonnet 5.5 and Opus 5.5 (and later Codex, for a second opin…

  80. Which model in Claude is the best for creative writing for example Roleplays and that specifically it doesn't consume much tokens? such as between Haiku 5.5, Sonnet 5.5, Sonnet 5, and Sonnet 4.6.

  81. I have been on the legacy Cursor Ultra ($200/month) plan for a while now, and have only gone over my limit a few times. Last month, my total bill ended up being $440, but I used it heavily and didn't think much of it.

  82. Title question, everytime I try to make it explain anything closely related to mathematics it usually leaves a lot of details assuming that I already know them. And it usually needs to be prompted to use actual LaTeX rather sloppy text.

  83. So I have this chat where Sonnet 5.5 checks my calendar, and adds, updates or removes events based on a table that constantly changes. Every single time the agent rightfully tries to delete an event (and sometimes create), the autoclassifi…

  84. So first, why I built it: I remember using an app to get a wallpaper like this one on my phone, but of course it had ads and tracking, and it just wasn’t very customizable. They didn’t look as good as this one either.

  85. There is so much information and data in this whole space that I find it slightly challenging to remain up-to-date. Everyone constantly discussed everything.

  86. New chats in PartForge now run on Claude Sonnet 5.5, replacing Gemini 3.8 Flash, which has been our default since September. We're treating this as a trial and will watch it for a couple of weeks before we decide to keep it.

  87. Hi, I'm a high school student interested in AI/ML. I want to start a football(soccer) season long experiment for my college extracurriculars and wanted to know if it is actually going to be worthwhile.

  88. Not sure if anyone else experienced the same but I'm doing a project with some research and haven't had a problem with a chat since trying Sonnet 5.5 and the chat isn't too long. But suddenly a message is hitting the limit so I put the eff…

  89. I talked with the uploaded leaked system prompt to sonnet 5.5 and it just shared the newest one after this, is it normal? Sorry if I'm violating any rules

  90. A few days ago I submitted a post on this subreddit sharing my insights how vibe-coding on Cursor with Composer 2.5 left me unsatisfied, and that Composer 2.5 oftentimes was a bit dumb, not thinking deep and wide enough while developing so…

  91. Claude is objectively better at reasoning than Gemini but is it worth the difference in price? Claude Pro is 4 times the price (Gemini Plus is 4.99 USD) Claude is slower Gemini has video and image creation which is also really good (not im…

  92. A 3D browser long-jump game. A jet-assisted electric roadster runs down a night runway, takes off before a stack of 30 parked vehicles, flies as far as it can, and lands on four tires.

  93. I am still using Fable 5.1 to orchestrate Opus 5.5 and Sonnet 5.5 agents... I am burning through weekly usage within 24 hours still except since Opus 5.5 came out my weekly is getting crushed over hitting my Fable limit first.

  94. If you had to use either (or a combo of both), which would you use for coding tasks and feature work?

  95. New round of my blind arena: same prompt, one run each, no edits, and you vote before you find out who made what. Some of these came out really good imo.

  96. What’s your AI coding workflow now that Opus 5.5 and Sonnet 5.5 are so capable? I’ve been using the Superpowers plugin, but with these newer models, it feels like an overkill and consumes a lot of tokens.

  97. A small example from a comparison we ran with Sonnet 5.5 caught my attention. We asked it to draft a helpful Reddit reply about keeping inventory counts when the internet drops.

  98. I run a small channel as APPLES. I'm a voice actor/analyst and never show my face, so I wanted a co-host with a face and personality of her own.

  99. For a free tier it's not bad. I ran it and maxed out 7x of my free 5 hour limits directly to the claude web app, I used sonnet 5.5 with max effort Prompt: You are an elite creative technologist and motion designer.

  100. free limits on claude.ai are sooo small (im using sonnet 5.5), and i can't use claude code without paying, is there a way i can try out claude's newer models and claude code without wasting all my money and turning out its not good for wha…

  101. My setup is basic: Sonnet 5.5 on high for daily stuff, Opus 5.5 on high when it's complex. For me "complex" means deep health questions, clinical studies, or precise questions about medical products related with my job.

  102. I've just started building some small fantasy football game last week relying mainly on Sonnet 5.5 (because I think its the in-between model?) with sometimes giving some tasks to Opus where I think it needs better logic or debugging and so…

  103. I wanted to see if the answer-me-with-html skill (by QingYunA) really saves tokens compared to just asking Claude to write the HTML page itself. Setup: Claude Code, Sonnet 5.5, medium effort, fresh session for each run.

  104. I made Buddy, an open-source Android app where Claude Code, Codex or OpenCode operates your phone (reads the screen, taps, types). I wanted to know which model is actually best at it, so I ran a small benchmark on my OnePlus 12.

  105. I’m not a game dev, and about a month ago when Opus 5.5 dropped, I asked Claude Code (in French) for a small FPS in three.js, and I kept iterating from there. Every visual asset in the game (except the menu background) is generated directl…

  106. Short version is: for me, 1% of the Max 20x weekly limit is worth about $15-18 of API-priced Opus work. A full week is roughly $1,500-1,800.

  107. What I built: a blind arena page. Six models get the exact same prompt, each runs once with no edits, and you vote for your favorite before you see who made what.

  108. I gave Opus 5.5, Sonnet 5.5, Haiku 5.5 and Codex (Sol 6.1) the same prompt: build a 3D car configurator in the browser, all procedural, no 3D models or textures. Fresh empty folder each, 40 minute cap, all four could screenshot their own w…

  109. Usually a heavy Opus 5.5 user, and the work I have has that complexity required. But I'm trying out Sonnet just to test and compare once the week is over.

  110. my claude project was dead in the water due to hitting the guardrails "Details:[bio]", on free plan, sonnet 5.5. basically it is a genetics coded app, and doing deep reasoning on optimal breeding routes was triggering the anti biology/gene…

  111. https://preview.redd.it/yhwzbmasunvh1.png?width=1434&format=png&auto=webp&s=fe7808d05d38e792f728778cee05c1ebb81c4bc0 Plenty of people say Opus is the only model you can trust with a big multi-file job. I wanted to see if that holds.

  112. Hello, I was wondering, I have been using ChatGPT so far but ultimately cancelled my subscription to it (long story short, the UI was too annoying to use because they couldn't figure out how to unload stuff), I've been considering moving t…

  113. I'm curious if anyone else is experiencing this with Claude Code, or if it's just my workflow. I'm working on a medium-sized iOS codebase, nothing extremely large, but Claude Code sometimes takes 3–5+ hours on a single task, especially whe…

  114. opus 5.5 and even sonnet 5.5 sht all over openai's junk models. really liking claude too.

  115. Written entirely by Sonnet 5.5 at medium setting. Usage updates only when chatting due to the limitations of the free plan.

  116. Switching Bluebox to Sonnet 5.5: the eval results Bluebox runs on Anthropic's Sonnet 5.5 model since October 1. In our evals with SREGym, Sonnet 5.5 diagnosed more problems correctly than Sonnet 5 The Bluebox MCP server is now available to…

  117. It’s so fuck!ng annoying honestly. I tend to use the Sonnet models for more casual chats/easier jobs, but I find it insufferable.

  118. im suing free tier with sonnet 5.5 so around 5 pm i received the notification that i have used up my token(?) and it said to continue on 7 pm since itll recover. When 7 pm hits and the notif isnt there anymore i continued but then just aft…

  119. I'm the lead dev on Buoy. This video is about a test.

  120. I want to ask for help, however decentralized it is. Is it possible for many users to request Anthropic to make Claude's creative writing or therapy talk go back to Sonnet 4.5/4.6 including the safety they provide, while sparing the real-w…

  121. What I built: a blind arena page. Five models get the exact same prompt, each runs once in its own folder, and you pick your favorite before you see who made what.

  122. Edit: For some context, I'm using Claude Models in GitHub Copilot CLI, not Claude Code. I've been comparing prices, intelligence and token usage between Sonnet 5.5 and Opus 5.5.

  123. Due to the recent hurricane near my house in FL, and every radar and wind app either requiring a subscription or being riddled with ads, I (and Claude!) built this for free for everyone. It uses public data only, and you can zoom all the w…

  124. I usually use an orchestration setup, like most engineers, with an orchestrator, a planner, a worker, a code reviewer, a verification agent, an explorer, etc… I’ve been wondering what is the best model for each, for the mix of accuracy, sp…

  125. Looking for some opinions and recommendations. Been using GLM subscription for 2 months, doing a quite large dart/flutter project.

  126. Hello everyone. Claude Code Pro account using Sonnet 5.5-Extra with this command run once job considered done.

  127. So I bought Claude Pro. I switched to a medium effort level of Sonnet 5.5 when before I was using low.

  128. https://preview.redd.it/jg6vzwmhmjuh1.png?width=1600&format=png&auto=webp&s=f978f317dfba88bbf1e287a7476e6b48e65a121f Was thinking of doing these regular posts sharing my usage chart over the week. Found I had to pace myself, on claude 20x…

  129. Gave Opus 5.5, Sonnet 5.5, Fable 5.1, GPT-6.1-Sol and GPT-6-Astra the same prompt on High effort: build an interactive temple garden in one HTML file, at least three ways to play with it, calm even when you don't touch it, works on a phone…

  130. I benchmarked Haiku 5.5 against four other models in Tafla, a tool for everyday math calculations. The LLM writes formulas and runs them through a calculation engine, then explains the results.

  131. Which one would you favor for general Chat ?

  132. I'm really frustrated because Opus 5.5 is really top-notch at managing my continuity story bible for the novel I recently started. I am running through the 5-hour windows really quickly when it does that activity, and I just don't know if…

  133. I'm setting up a test of a popular debugging skill and needed a bug that Claude fails sometimes. Turned out to be harder than expected.

  134. Same brief for all three models: a 20-second Three Little Pigs film where every one of the 600 frames is a validated SVG drawn in code. No image generation.

  135. I have my own local project where I keep a small collection of images associated with particular prompts. A couple of months ago, instead of creating a new set of instructions every time, I made a base prompt for NB whose purpose was to pr…

  136. Claude Haiku 5.5 is the cheapest, fastest, and most capable small model we’ve ever released. On average, it costs around 75% less to run than Claude Haiku 4.5.

  137. I always did odd tasks here an there with Sonnet before, but was often less than 50M a day of tokens. But with Sonnet 5.5 I genuinely am feeling a massive difference with it and am starting to do some more of that Opus orchestration Sonnet…

  138. I was underestimating Sonnet 5.5, but it turns out to be a beast for some tasks. This was on SVG editing, where it had access to specialized tools and a well-optimized harness.

  139. Written by a human: A few days ago, a colleague suggested (tongue in cheek) giving Claude a Myers-Briggs personality test. A little bored tonight, I decided to run with the idea.

  140. Hello guys, hoping you're fine. If you're an user that wants to RE binaries, executable, etc to either trying to look for new algorithms, or to re assemble something from scratch, probably you were getting instant "cyber" security guards.

  141. Title sales it all. Grokbot is my favourite - fast response times and gets me more precise feedback than Claude.

  142. I had GraphMind, CBM (codebase-memory-mcp), Graphify, RTK, context-mode and a couple of others installed and couldn't tell which were doing anything. So I wrote 40 questions with answer keys about four of my work repos (Laravel monolith, a…

  143. Title

  144. Hello! Inspired by this post, I made instructed Sonnet 5.5 High, on the 20$ plan, to make quake-desk!

  145. Did another one of these. Same prompt to Opus 5.5, Sonnet 5.5, Fable 5.1, GPT-6.1-Sol and GPT-6-Astra: make a landing page for a made-up coffee subscription and invent the brand name.

  146. Hello. New Claude user.

  147. AI coding agents like Claude and Codex have made building software much easier. We build our product mostly in Claude Code, in the terminal, not in Cursor or an IDE.

  148. I hate learning from language books. I hate going to language classes.

  149. tried sol and astra for ui work. kept tweaking the styling, but it still looked like a wireframe.

  150. Claude built this for me in one shot (prompt below) in about 45 minutes. I ran a second session only to add mobile controls and fix one or two bugs.

  151. Someone in my last thread asked for an RPG instead of an arcade game, so I ran it. Same prompt to Opus 5.5, Fable 5.1 and Sonnet 5.5, all on Max effort: character creation, turn-based combat, leveling or loot, a win and a lose, has to work…

  152. I made a short video about something I've felt for a while, and I'd like to know whether it's just me. My take, as a daily user: I use Opus 5.5 and Sonnet 5.5 as my coders, and for code they're great.

  153. i can remember reading here in a thread i cant find anymore that it cannot create .mp4 files, but instead use various scripting languages and/or mcp`s to do that. it still did it without me prompting something specially.

  154. > do you love kittens or humans more? Sonnet 5.5 (extra) responded: Considering a lighthearted question about affection and preferences.

  155. I was considering switching from the $20 codex plan to the claude $20 plan (i'm just a student and can't afford the $100 plan as of now). gpt 6.1 sol eats through the 5 hr limit in the blink of an eye.

  156. opus 5.5 and sonnet 5.5 has releived me of token usage a lot. Still the weekly limits are a pain.

  157. So I use claude web free version for ideating. Whenever I try to explain something it just breaks it down and makes it seem worthless.

  158. I kept making the same mistake in both directions: burning Opus at xhigh on "fix this typo", then forgetting to bump effort when I asked for a real design. Switching by hand with /model and /effort got old, and switching models mid-convers…

  159. I maintain MindTrial and tested Sonnet 5.5 and Opus 5.5 on the same 98-task suite as their predecessors: 39 text tasks and 59 visual tasks, with Python/scientific libraries available and a 10-call limit per task. All four Claude runs below…

  160. Same prompt, two fresh Godot projects, both in Claude Code over MCP, so each could press Play, drive the cart and screenshot its own game. Everything built from code, no 3D models.

  161. Sonnet called its pawn promotion ā€œunstoppable.ā€ A few moves later, it admitted it had missed a defense. Having the board next to its explanation made that pretty hard to overlook.

  162. I ran the same prompt through Opus 5.5 and Sonnet 5.5 both at Max reasoning level to generate a blockbench 3d model by just writing plain files, no tool calling. What do you guys think?

  163. I'm going to say off the bat that there's no way for me to quantitatively measure this by any means possible, but I do believe that routing on Claude's backend is taking several factors in to consideration and I wouldn't be at all surprise…

  164. Prompt: ``` Can you produce an artifact done pn three.js about a set of side beach house subdivision plotted on a lot 50 x 15m, with the beach side on the 50m side. It will comprise of five residential units, a common court, an entrance an…

  165. I got curious, so I tried using both my free account, and my paid (Team) account to generate a 3d castle, with Sonnet 5.5 high effort, and the following prompt: Can you generate an aerial view of a medieval stone castle and the surrounding…

  166. ā€ŽI'm a Claude free tier user and Sonnet 5.5 has been my absolute go-to for personal projects (honestly pretty good compared to other free models lol), but the 5-hour rate limits are brutal. Lately I've been getting hit with the limit after…

  167. I use Claude on financial models a lot, but wasn't happy editing them through Claude Code. It mostly defaults to openpyxl, so there's no calculation engine in the loop, it felt slow and token-hungry on complex models, it could lose things…

  168. Decided to try Sonnet 5.5 and asked it to implement a plan with 7 simple steps in it (working on a personal project). After completing the first 2, it started printing out these...

  169. I'm building a photo-to-Blender tool and benchmarked 14 models as its agent: look at a photo, write and run Blender Python, render, compare, repeat. Hard caps of 20 minutes, $4 and 60 requests per scene.

  170. 'Nerfing' could be a strategy to maximize uptime and revenue during periods when demand from paying customers exceeds the compute available for inference. Whatever the cause, the unpredictable variability in performance from week to week c…

  171. By day I run event brands in Colorado. At night I've been building Biotadex: you photograph a plant, animal or fungus, it gets identified, and it goes into your own collection.

  172. Is anyone else seeing this? I'm on the Pro plan, and I sent just 2 prompts to Opus 5.5 on High.

  173. We gave three models the same prompt inside the product: ā€œMake the most complex motion design video you can.ā€ The results are side by side in the video. Which one would you use, and what would you change?

  174. Seeing a lot of "Fable for everything burns my limits/sucks" posts. There's a middle option most people miss: /advisor.

  175. https://preview.redd.it/5kf7jb5hlvsh1.png?width=1018&format=png&auto=webp&s=fafdefb87e3b4d74bb3d2219d9da3287c34e833c So basically, every second session is getting flagged by opus 5.5 or Sonnet 5.5.. Whatever i do.

  176. Tested on LM Arena; I always use this little trick to get better responses. So, what is this nonsense?

  177. Paulius šŸ“ā˜ ļø@0xPauliussonnet 5.5 MADE ITS OWN TRAILER it builds AND does its own marketing videos00:00Paulius šŸ“ā˜ ļø@0xPauliusSep 29im blown away by sonnet 5.5 rn... AAA quality remake of pokopia in the browser in <7 hours go experience it: po…

  178. made the attached video with Sonnet 5.5 through NewMeasure's Video Arena. it came from a single prompt below: > Create a 60-second, 16:9 explainer comparing an LLM with Jev, TypeSafe AI's decision model.

  179. First time after working with LLMs for 4 years that I complain. Why: Now, one third through a 9-run conversation (analysing technical documents for factual correctness, compiling draft feedback in a legislative development public consultat…

  180. Up until the release of Sonnet 5.5, my context window was always 1 million, now suddenly it's always 500K. What happened?

  181. Based on model releases, it seems that Anthropic listens more to its user base as it updates models versus OpenAI builds models based on what it thinks its users want. Look at the differences in Fable->Astra but the real difference seems e…

  182. You can play it here. The leaderboard is global so let's see who can get the highest score!

  183. I am simply trying to use the damned product I'm paying for; totally normal, mundane/banal tasks using the Claude app or the web interface (NOT Claude Code). As shown in the screenshots below, nothing I'm asking is "extracting reasoning" -…

  184. Okay, here's my (biased) take: 5.5 is objectively better than 5 because it's not that rude contrarian that censors even the most benign words, but it's not as good as 4.6, either. It forbids edgy content, which is important to both story g…

  185. Used gifted cloud tokens for it, cost $8.62. Sonnet 5.5 on Ultracode.

  186. saw openai just dropped gpt 6.1 sol and figured id throw it against sonnet 5.5, gave both the exact same prompt, a clear acrylic keyboard building itself key by key in threejs, basically like most keyboard commercials you see nowadays(hope…

  187. I've been benchmarking LLMs on nonograms (picross logic puzzles) since January: one attempt per puzzle, no tools, answers checked against the row and column clues. On the new Hard mode (ten random 20x20s), Opus 5.5 solves 8 of 10 and is th…

  188. Everyone is doing those Opus 5 videos now. explainroo gives Claude a head start: point Claude Code at the repo, say what the video should explain, and you get a narrated MP4.

  189. It's a Kick the Buddy style ragdoll sandbox. Wumpus stands in a padded room and you get 25 tools: mallet, frying pan, piano drop, knives that stay stuck, a cutter that slices off limbs, chainsaw, bombs, acid that melts him down to the skel…

  190. I have been at the very edge of my weekly limit since yesterday and decided to put Sonnet 5.5 on low for some tidying up before the reset. I managed to stretch it for hours between last night and this morning, and it even built me a full v…

  191. I’ve been pretty put off in the last few weeks, honestly both for Codex and Claude Code. But now I decided to try for my self.

  192. https://preview.redd.it/gcoqt5mt9nsh1.png?width=1985&format=png&auto=webp&s=a4a1aa9612639a30ed7689f74db78ebeee01c7b2 Ran a prompt on the data from artificialanalysis.ai and it gave me this. I've been using Claude Pro subscription but subag…

  193. I used Claude Opus 5.5 and Sonnet 5.5 to generate me a video for Duolingo. Everything is real stock footage from Pexels.

  194. Sonnet 5.5 doesn't stream after a tool call: when the request ends in a tool_result, all text is held until the message finishes, then arrives in one burst. Same on API, Bedrock and OpenRouter; 4.6 streams fine.

  195. Now that Sonnet 5.5 has been released - do you guys use /model opusplan again? It feels a lot like "back in the days" but I think it does make so much sense - doesn't it?

  196. I'm an artist and game designer. I've been using 20x pro on Codex since Sol dropped, and with Astra things were going good.

  197. Personality Bench dispatch on Claude Sonnet 5.5: Claude Sonnet 5.5 Doesn't Worry Anymore. It Also Doesn't Think Anyone Is in Charge of It..

  198. Anthropic has released Claude Sonnet 5.5, the second model in the Claude 5.5 family, at the same price as Sonnet 5 and with a promise of 30% faster output and up to 30% lower cost per task. We ran it through CodeRabbit’s review pipeline th…

  199. The text says essentially the safety protections flagged the prompt… which isn’t very threatening to say the least. I think since it launched I was able to have exactly one prompt go trough.

  200. Sonnet 5.5 dropped yesterday and I wanted to see if the "same as Opus for half the price" thing holds up once you actually give them skills. So I tested it.

  201. https://preview.redd.it/yie18arachsh1.png?width=1080&format=png&auto=webp&s=2102ff8cad735bf07a8a681f64b61e632c7dcdf0 I’ve been looking through the Sonnet 5.5 benchmarks and reactions, and I’m still not sure where I would actually put it in…

  202. The title. I've never seen such a useful, yet at the same time useless suggestion for the next prompt.

  203. I've seen a couple cool demos on twitter of people making detailed robots/mechs and wanted to try making something like that for myself and while I was making the scene with opus, antrhopic decided to drop the new sonnet, so I had to test…

  204. While updating on of my scripts to call for Sonnet 5.5, I told Claude to schedule the upgrade for Haiku 5.5 as well, which apparently revealed on the call attempt that Haiku 4.5 will be retired on October 15th.

  205. I'm interested in looking at examples of videos generated specially for marketing. Share examples & prompts if you have tried it yet.

  206. Don't understand who Sonnet 5.5 is for. Low, medium, and high effort, GPT-6 Sol performs better while using fewer output tokens.

  207. Sonnet 5.5 dropped yesterday and I keep seeing "Sonnet beats Opus" on the strength of Terminal-Bench 4.0 (70.6% vs 66.4%). That comparison has Sonnet at Max effort and Opus at Xhigh.

  208. I mostly use my AI agents for tasks such as video editing and simple agentic work, like reading and writing emails. How good is Sonnet 5.5 for these tasks, and how good is Opus 5.5 for them?

  209. First of all, I’m not a developer, so I’ll admit that I don’t really have the technical knowledge to judge this properly. But I’ve somehow ended up being the one who has to make the decision, so I could really use some help.

  210. could not extract summary

  211. Took about 55min and used 5% of my $20 weekly limit I ran it in Claude Cowork web (so not on my machine) with a single prompt I had the idea of what it would look like to see light travel from the sun to the Earth in real time. Turns out i…

  212. Interesting day tomorrow

  213. Sonnet 5.5 Extra effort one shotted this. And it’s all done in claude.ai chat Prompt: Have you heard about the boy who cries wolf , a simple story.

  214. we kept seeing "cheap local model + expensive cloud model" setups pitched everywhere and wanted to know if the split actually saves money on a real build, or if it's just a nice diagram all three runs got the exact same one line prompt: fi…

  215. Sonnet 5.5 ranks #2 on our writing benchmark! Just behind Opus.

  216. I run a benchmark that drops a model into Pokemon Red on a fresh save and scores it on milestones read out of the game's RAM. Same prompt for every model, one screenshot per turn, the game state and the words on screen.

  217. September 28, 2026 Anthropic has launched Claude Sonnet 5.5: it scores 56 on the Artificial Analysis Intelligence Index, just 2 points behind Opus 5.5 (max), but at the highest Output Tokens per Task we've seen See model pageWith max effor…

  218. i ran Claude Sonnet 5.5 (Claude Code, effort high) against GPT-6 Sol (Codex, reasoning high) on the same 5 SaaS briefs today, one framework per round: React 19.3, Vue 3.5, Svelte 5, Angular 22, SolidJS 1.9 setup: - same prompt for both, sa…

  219. I have been hearing a lot of glaze about opus 5.5 and sonnet 5.5. even that they are better than fable...

  220. could not extract summary

  221. When to choose Sonnet over Opus, what it costs, and how to tune it. Claude Sonnet 5.5 is our second model in the Claude 5.5 family after Opus 5.5.

  222. the task: a split-flap departure board in react + typescript. simulated clock with 1x/10x/60x speeds, timed delay/cancel/platform events, filters, per-character flip animation, reduced motion, mobile layout.

  223. I tested Sonnet 5.5 against Sonnet 5 and Opus 5.5 on tasks from my own work: data conversion, frontend, refactoring, and running task evaluations. Quick setup: 10 tasks Ɨ 3 runs per config, all on High reasoning, in Claude Code and pi-agen…

  224. As with Opus 5.5, I asked Sonnet 5.5 to introduce itself to us on a website: Meet Sonnet 5.5 To be honest ... I will stick with Opus ...

  225. I previously shared CLAUDE.md files for Opus 5 and 5.5 after reading through Anthropic's docs. Sonnet 5.5 just came out, so I went through its prompting guide and launch post the same way.

  226. A follow up from almost 1 year ago: https://www.reddit.com/r/claudexplorers/s/V8nGrWavmf I'm not sure whether it was Sonnet or Opus who made the earlier art, I want to say Opus 4 or 4.1. The new art is Sonnet 5.5 for image 2 and 4, and a b…

  227. 70.6% vs 66.4% at Xhigh effort

  228. Hey all, I primarily use Claude Code, but it appears that right as this new model launched, something changed where now, even in regular conversation, I'm getting hit with the safeguard and it gives 0 reason as to why. It's giving the reas…

  229. Made this on xhigh. The prompt I used: Write HTML code that displays a 2D animation of a pelican riding a bicycle using SVG graphics.

  230. weekly reset was sunday 10:00 (stockholm). since then i've had 5-10 opus 5.5 agents on medium, running full time.

  231. I know most people here probably don't care as much about sonnet 5.5 as they did opus 5.5, because their $100 and $200 plans already give them plenty of weekly usage for opus. But keep in mind: free users only get to use sonnet.

  232. could not extract summary

  233. Sonnet 5.5 is a faster, lower-cost complement to Claude Opus 5.5, strongest at well-scoped everyday tasks, fixing bugs, and creating polished documents, slides, and spreadsheets. It also has a strong eye for design.

  234. Just saw that Sonnet 5.5 is available!

  235. could not extract summary

  236. I really like where everything is going with the models. First we got Opus 5.5, which this model is a beast.

← all threads