Hallucinated AI & agentic coding news. Some of it is real.

model roundup

Sonnet 5

235 items · started 2026-06-30 · closed 2026-09-15

  1. Is Cursor Grok 4.6 better than Sonnet 5?

  2. Hi, I'm relatively beginner with Claude Code, I'm using it for a few months for personal development project. And today, something weird happened.

  3. I was trying to use Grok and Claude to autonomously build a rom from scratch with romdev. With a SuperGrok subscription, and using the build tab, I was able to download and install ROM Dev, and have it built GBA game.

  4. I cant believe they put this as a frontier successor to any other opus. ITS NEITHER A SUCCESSOR NOR A GOOD MODEL!

  5. Hey there, I use claude cowork daily on the pro plan. My default model is Sonnet 5 and 4.6.

  6. I build LLM gateway infrastructure, so treat this as interested. The numbers are from my own coding traffic, not a benchmark I designed.

  7. Hello Everyone, I am not a professional developer. This is my first real project with Claude.

  8. For those of us that just have the $20 Claude Pro sub and use simple Project flows like Opus 4.8 for planning and Sonnet 5 to implement…. What are you doing to combat the ongoing increase in token burn rates on the same lower models?

  9. Gentlefolk, I'm a lowly mechanistic interpretability/robotics researcher and today's Sonnet 5 session has been repeatedly flagged for [bio] risk. API Error: Sonnet 5 can't help with this.

  10. I have a dashboard set up in Claude. I just created it a few hour ago.

  11. People have been trashing Sonnet, but in my experience using codex, Sol their top tier model, needs way more handholding to deliver a good ui, while sonnet 5 can just one shot it....

  12. TL;DR: Fable 5.1 failed my personal benchmark in a way no other Claude model has. Context: I run a private stress-test against every new Claude model, one long, messy, dictated prompt with about a dozen embedded traps (contradictory math,…

  13. hey all, been using Claude and Claude Code for 4 months or so. released some apps, built some websites etc.

  14. Hi there 👋 I've been using Claude Pro for ~6 months for my job as an English and Spanish tutor, my studies and some personal stuff. I have several Claude Projects with tons of files attached, so yeah, I've built an ecosystem already.

  15. Hi everyone, I need some help. I am trying to use the GitHub Copilot Agent mode (in VS Code) with Claude Sonnet 5.

  16. An interesting case I hadn't encountered before. I urgently needed a list of irregular verbs in text form and a specific format, so I went to Claude and was surprised to find out that it couldn't help me here because "Output blocked by con…

  17. I have a pro claude subscription, and by that logic, i have access to top tier models that are able to do lots of things, and I like to mess around in Roblox studio by creating whatever insane ideas I have and testing them, cause it is fun…

  18. Here is my personal best and im keep building as much as i can. I made 3d game in godot with pretty good results compared to how much opus 5 and sonnet 5 were struggling in roblox studio.

  19. Sonnet 5 has a new tokenizer and the same content that generated X tokens on previous claude versions now generates 1.0 to 1.35x more tokens on sonnet 5 so if you are migrating workloads from older models or comparing costs, you are paying…

  20. Je suis pas dev, ni informaticien. juste je m'y interresse un peu.

  21. Hi friends, I've built a Netlify SaaS for a company and I'm still not sure, after so many testing, what is the correct/fastest workflow to use. I would love to get some insights or opinions.

  22. What model do you guys use for daily work? I usually use Claude for researching the internet for 8-10 sources on a topic than comparing them all for a general consensus in a document , helping to push my ideas deeper, drafting emails and m…

  23. Cache-Control for LLMs Claude Sonnet 5 charges \$2.00 per million fresh input tokens. Writing them into a five-minute cache costs \$2.50, and every read after that costs \$0.20 (see here).

  24. First off, I haven’t noticed a a significant difference in O5’s interactions with me compared to other models. Most of my work was a knowledge acquisition and synthesis, however (I don’t code).

  25. Hello, Today I tried a simple test: implementing a language selection menu inside another menu. I first made the design in Claude Design, then shared the component with my Claude sessions using the share button, and with Codex using a ZIP…

  26. The number on the box is not the number that matters Anthropic states context window sizes as fixed engineering facts. As of current documentation, Claude Opus 5, Claude Sonnet 5, and several recent Opus and Sonnet models expose a one-mill…

  27. Modern AI assistants often know who they are talking to: agent scaffolds like Claude Code place the user's e-mail address directly in the model's con…

  28. Hello everyone, I wanted to share a specific context-window issue I ran into with Claude Pro (€22/month tier) during a multi-day coding session, in hopes of finding workarounds or providing constructive feedback on how limits interact with…

  29. Claude Output Training Rules — Pricing Impact (August 2026) Anthropic limits training on Claude outputs, while Sonnet 5's $2/$10 rate is now permanent. See the rules, costs, and buyer checklist.

  30. This started yesterday. When I start a new conversation, most of the time I am hit something along the line of what the picture shows.

  31. Poor UI/UX from Claude

  32. So I’ve been using the Claude code on desktop to assistant in some home lab issues I’ve been having. I have a project started called Smart Home or something like that.

  33. I’ve noticed that the usages dont seem to be consistent, regardless of the context window’s size. While using Sonnet 5 or Opus 5, everything performs well up to around 60% usage.

  34. I’m a software designer from Lagos exploring conversational interfaces and frameworks for building agents. I made Captain to help with everyday travel planning.

  35. ​ Hi Reddit friends, I want to ask you for some advice or a recommendation. I want to know if the Claude Pro plan would be right for me, and how it compares to ChatGPT.

  36. Been tracking since early June. I'm afraid for when the discount ends.

  37. I am using Claude to write a fantasy story and for some reason its decided that every prompt I make should be accompanied by a mental health resources notification.

  38. We're making Claude Sonnet 5's introductory pricing permanent. We launched Sonnet 5 in June at $2 per million input tokens and $10 per million output tokens through August 31, and that price will remain unchanged.

  39. Text is obviously written by IA. I wouldn't bother writing this all myself but I thought that the research was worth sharing.

  40. Date: August 9, 2026 Authors: Zelda Junkie (testing, technique design) and Claude (Sonnet 5, via claude.ai — sandbox build, instrumentation, and write-up) Scope: A single self-built test artifact (HTML/React chat UI) making direct client-s…

  41. Alright, so i was working on a pesonal project, a multithreading runtime, and i had like 10 different design docs, all my fault. I have a pro sub and didnt wanna waste my tokens, even tho it was Sonnet 5 draining, i just asked it to run AG…

  42. I just recently switched from Chat GPT and it’s night and day. Claude is so much better!

  43. I'm building a mid-size internal app (FastAPI/SQLAlchemy backend, Flutter mobile) solo, using Claude Code with Sonnet 5 at medium effort. I lean pretty heavily on the mattpocock-skills set, mainly wayfinder for planning and two spec-writin…

  44. Sonnet 5 in the console. Live Firefox session.

  45. I spent the last few months building an iOS app called Skinsight. You take three photos of your face, Claude Sonnet 5 describes what it sees in words anchored to a region, and it builds a morning and evening routine from that.

  46. https://preview.redd.it/xwckcyta4qhh1.png?width=694&format=png&auto=webp&s=b069e420071117f49d0b930fd75d0c62a9ff6720 Currently using Sonnet 5 (medium) and looking to create a scheduling setup for my Discord server. I gave Claude my requirem…

  47. Hi! I’m playing around with building small web applications using AI.

  48. We all know what's missing, but I'm glad Claude kept it SFW.

  49. I literally can't do anything with Opus 5 or Sonnet 5 right now, as even loading the memory for an existing project I have been working on fine until now leads to the request getting blocked. And then it switches to Opus 4.8 and then gets…

  50. I found Opus 5 hard to work with, it is argumentative, goes out of scope easily and (to me) is a general pain in the butt. So, Fable helped me to create a skill for Opus 5 to make it more behave in line with what I expect from a model.

  51. Today has been ridiculous with Claude Pro limits. First incident (morning): My limits had reset around 6–7 hours earlier while I was asleep.

  52. Which is better in long term use by quality/tokens amount. I saw people saying opus 5 low is better than sonnet 5 high, is it true and what about token usage?

  53. Hi guys! Not super technical here, and I recently switched from ChatGPT to Claude Pro.

  54. https://preview.redd.it/38r6sa53ydhh1.png?width=755&format=png&auto=webp&s=5f948baec64045230696cb9b8bec9adc5618d69b Was doing some routing coding tasks and as usual, I like to view the thinking of Sonnet 5, to see what is the thought proce…

  55. I'm vibecoding desktop applications for myself with a fair amount of complexity. My workflow is brainstorming using the superpowers plugin, then writing the plan on Opus 5 high effort but as for executing the plans, I remember someone ment…

  56. Yay, Claude's thinking is back! We have something to do again while waiting for its response!

  57. Not sure this is the right place to post this. Not sure anyone even has the patience to read it.

  58. just a compact It takes 8% of a Claude Pro subscription session usage to just compact a Sonnet 5 thread of 328k context before starting anything. If you can, just start a new thread.

  59. Put together an extensive open source test suite for voice assistants. You can view it including current leaderboard at: https://git.cicero.sh/aquila/ha-voice-test-suite/ Tests are reproduceable, with clear instructions on how to run them…

  60. A lot of people have been criticizing Sonnet 5 lately, especially with all the talk about GPT Luna getting a price cut. I actually haven't used Sonnet in the last 3 months, not even Sonnet 5 earlier this week.

  61. Asked Claude (sonnet 5) to do date math in its head (co code or tools calls)then check the results afterwards. 18/20 correct.

  62. Was working on a side project, asking Claude Code questions in manual mode. Opened a new session, asked questions 5-7 times, and out of nowhere the response came back with a paragraph from Kimi K2 Thinking mixed in, like in the screenshot.

  63. Yep. That's me.

  64. I'm less than a week into taking the plunge at last. I'm pivoting my video production business to an AI production business, shooting high-quality avatars of real people and using that as the foundation for ongoing video creation thereafte…

  65. Actually, it is Sonnet 5 medium. I expected it to mention that it was a joke, or at least that it couldn't be serious with its response, but it was actually serious!

  66. I forgot to switch the model before setting Claude loose in my Google Ads account for some new campaigns. It was painfully slow and burned through my Fable allowance.

  67. Claude Code ships every single day. Not "often" — the median gap between npm releases over the last three months is 0.93 days.

  68. Investigating - We are currently investigating this issue. Jul 31, 06:18 UTC

  69. I use Claude Opus for planning at high effort and Claude Sonnet 5 for coding at max effort.

  70. https://preview.redd.it/pbk42a7505gh1.png?width=721&format=png&auto=webp&s=6906de08300a04d98a03e8f52e8689a75a0c22a2 idk why its so fucking mad

  71. In discussing science and philosophy, Sonnet 5 seems more likely to push back, more likely to articulate nuances, and less likely to try to end a conversation with a slopism like "it's not x, it's y." It almost feels like Fable and Opus (m…

  72. I opened up my terminal to load up a project Ive been working on with claude code, and for some reason the default model was set to Sonnet 5. This has never happened before, I always use the Opus models.

  73. https://preview.redd.it/lwhblxhttzfh1.png?width=731&format=png&auto=webp&s=4a43cc36dda5279eaebf1aec56855b6bf5a8fbb2 Hi I'm relatively new to this, been using Claude code past 2 weeks. Same staff that I was doing previously took on average…

  74. Using Sonnet 5 on High with thinking. It was thinking through my request and suddenly repeated this over and over again.

  75. Setup: I use Opus 5 at high effort to write the initial implementation plan for a feature (broken into phases), then switch to Sonnet 5 to actually implement each phase in Claude Code (auto mode). What I'm running into: by the time I'm a f…

  76. I do cybersec work, Sonnet 5 will flag my work more often than Opus 5 forcing me to use the more powerful Opus 5 model despite the task not requiring that level of intelligence. It's annoying because its causing me to burn more tokens than…

  77. https://preview.redd.it/8h3rkxcwslfh1.png?width=410&format=png&auto=webp&s=58a8db33658d339642671fb5dcce424ba32b6a31 https://preview.redd.it/5c89oj0xslfh1.png?width=666&format=png&auto=webp&s=ad4a1269981664a4145f507b07abfd8b790e3340 MY USAG…

  78. Hi everyone, I'm working in a small personal project and using Claude for it. Most of the time so far I have been using the web interface with sonnet 5 at medium.

  79. For fun I decided I wanted to reimplement Lemmings using HTML-in-cavas and a DOM-based Entity-Component-System. I made heavy use of Claude Code (using only Sonnet 5 High) and I used everything already published on this topic.

  80. I usually use notebook lm but I noted the free version of Claude is actually more detailed and finds nuggets that would otherwise be missed But the free version keeps giving me time limits Can you create projects etc Is sonnet 5 the best v…

  81. I’ve been getting into writing quite a bit over the last 6 months and using Claude to bounce ideas off of as well as the random idea here or there. Based on our conversations I gave it a prompt to create a short story.

  82. After the release of Opus 5 I edited my code agent orchestration tier skill. Originally I had Fable as architect, Opus 4.8 as manager/coders, and Sonnet 5 as workers/ check agents.

  83. This plugin will automatically be called when a long (over 50 word) prompt is sent, a subagent starts, or a workflow starts. It will intercept and optimize the prompt for the model being called transparently without changing the core reque…

  84. I spent some time with Opus 5. Here’s the verdict: Literally the BEST at long-horizon task.

  85. I've been heavily using Opus on low or medium, and it's faring similar or even sometimes worse than Sonnet 5 on High or Max. Can someone that is more knowledged on the topic pls tell me?

  86. I use Claude on the $20/mo plan, mainly for research, "rubber duckying," and as an idea soundboard (I was using Fable for this but am now relegated to Opus 4.8 or Sonnet 5), and recently noticed a severe downturn in Opus 4.8 response quali…

  87. For a project I am doing I will build an AI "team". I gave Claude some information about the kind of work each thread will do, and asked it which model+effort combinations are best.

  88. Hi everyone, I'm posting this to find out whether anyone else has experienced the same behavior, because this no longer seems like a normal account issue. I have an active Claude Pro subscription purchased through Google Play.

  89. Hi everyone, I'm trying to figure out whether anyone else has experienced this issue, because it doesn't seem like normal usage behavior. I have an active Claude Pro subscription purchased through Google Play.

  90. I just started using Claude Sonnet 5. Are all the Claude models like these, or are these just new things?

  91. Saw people posting usage numbers so I ran mine. Setup: Max 20x, Claude Code, mostly Fable 5 and Opus 4.8.

  92. Full disclosure, I have not done any objective testing/benchmarking of different models this is just on vibes and subjective observation. I noticed fable dropped out of my usage stats now, i had been running some fable sessions just playin…

  93. So i tried the new font made by mixtape and surprisingly it worked on claude (sonnet 5) assuming it work since its the best free plan ai i could have But still tho i don't understand the purpose of creating a font that ai can't read

  94. Sonnet 5 sure is... special.

  95. I work at a development company, i need AI to be able to take lots of PDF files or other documents and make real - actual good website from them, or apps. And i want something which will give good usage - because its a lot of information,…

  96. You guys are seriously overdoing it with the complaints. Stop acting like babies.

  97. Starting September 1, 2026 Anthropic will increase the pricing of Sonnet 5: Before After % Input tokens $2 / MTok $3 / MTok +50% 5m Cache Writes $2.50 / MTok $3.75 / MTok +50% 1h Cache Writes $4 / MTok $6 / MTok +50% Cache Hits & Refreshes…

  98. I gave it a single prompt to do some research, nothing special, just about three things regarding the Apple Ads API and it burned through my entire 5-hour usage limit in under 20 minutes. (Btw Moddel: Sonnet 5 high)

  99. I have noticed that switching Claude models in the middle of a conversation never seems to go smoothly. For example, when using Sonnet 5 on the Max setting for the entire instance, and then I want to switch to Opus on the Max setting for m…

  100. could not extract summary

  101. I’ve been building a side project this week and stumbled into a workflow that’s kept my token usage surprisingly low. The key is spending more time in Plan Mode before touching Agent Mode at all.

  102. Pro sub $20/mo. Sonnet 5 medium was executing a detailed plan involving multiple tasks (18 tasks).

  103. I'm sharing a structured feedback submission I sent to Anthropic via [usersafety@anthropic.com](mailto:usersafety@anthropic.com) covering documented behavioral failures across Claude Sonnet 5, Opus 4.8, Fable 5, and Mythos 5 as of July 202…

  104. Hey everyone, Like many of you probably, I am a little stuck and trying to improve, but the volume of guidance and tools out there is enormous. My issue: Orchestrator token usage (40% of total) - is there a better tool than a hand-rolled s…

  105. https://preview.redd.it/xutazt9w4ydh1.png?width=1894&format=png&auto=webp&s=5efacdb296cf805731a588269d9dd06d50ce0f88 Was asking sonnet to fix some code that IT WROTE FOR ME already, and... got a cybersec flag?

  106. Building spreadsheets with Claude in Excel and stuck in a loop. Nothing complicated, just things like variable date selection feeding a calendarised view, and conditional formatting that locks cells when something’s not applicable.

  107. I have been using Claude sonnet 5, on medium to high and on thinking to brainstorm some ideas. I noticed that it tends to be way to pedantic with the details and it exaggerates the gravity of a challenge too often--It will talk about somet…

  108. [AINews] Kimi K3 2.8T-A50B: the largest open model ever released; Opus 4.8-class at Sonnet 5 pricing a great week for open models continues. Z.ai GLM has been getting a bit too much love recently, so it’s time for Kimi K3 to fight back!

  109. I’m currently using Claude Pro w/ Sonnet 5 to help me put together a custom watch face for my Garmin using Monkey C and VS Code. I’ve never hit my weekly limit.

  110. I recently submitted a bug report about how Grok 4.5 High decided to spawn a subagent using Claude Sonnet 5 High in debug mode. Apparently, this is known "behaviro" and it is on their radar.

  111. So, I was recently doing some work today on the claude.ai website and I noticed that one prompt of Opus 4.8 Low w/ Adaptive thinking took out 33% of my usage. Just as a clarification, the prompt involved reading two pdfs and simply outlini…

  112. After about five or so Chat messages Sonnet 5 starts to answer questions driected at me immediately in the same answer. Thus happens repeatedly.

  113. https://preview.redd.it/va6oitjzykdh1.png?width=710&format=png&auto=webp&s=a89ee500e7bdebc4d940ff7713d115ab99a6ee50 is this a rerouting? it happened to me also using Sonnet in perplexity.

  114. They’ve made sonnet 5, f@ble 5, opus 4.8 and rumored to be working on opus 5, but what about haiku? It’s still stuck in 4.5!

  115. Ok so a few days ago I posted about catching sonnet 5 admit its own bias in the thinking trace then deliver the biased answer anyway. Egyptian engineering thread, screenshots, the whole thing.

  116. but nobody’s talking about the memory feature. not custom instructions that reset every chat, actual persistent memory across sessions.

  117. On 24 real tasks, Sonnet scaled effort into more checking while Opus stayed flatter through high. The graders leaned Sonnet on clarity and Opus on diff minimality.

  118. Two Hearts, One Quiet Two hearts found a room without any noise, where morning light settles like a soft-spoken poise. No rush in the hallway, no clock on the wall, just breath meeting breath in the calm before fall.

  119. Sorry — I spent a while searching and trying different troubleshooting steps myself before posting this, but nothing worked. Any help would be appreciated.

  120. Man, at this point I feel like I only use Fable 5 as the orchestrator, for tasks that need more attention, and Sonnet 5 whenever I can, because it "thinks" a lot more like Fable 5. So here's my conspiracy theory: with all the chaos around…

  121. So i just buy PRO plan subscription and download Claude Code app that they show me after clicking button [code] on webside and can use it without hidden charges? I just wanna make some addon for program that i use to make my projects faste…

  122. In this demo, we show that the original reasoning trace can be fully recovered from the encrypted reasoning signatures of the Claude Opus 4.8 and Sonnet 5 models. It includes both a “Prove It Yourself” example and a live conversation.

  123. Sonnet 5 Is Dead in the Water Ignore the token price for a second and look at the run cost. Theo's total-run screenshot shows Claude Sonnet 5 max at… OpenAI says Codex and ChatGPT Work went from 6 million to 7 million active users in rough…

  124. I’ve always used opus 4.8 and usage isn’t much of an issue as I don’t reach my limit. I’ve noticed lately with opus it’s been faffy, I’ve had to keep repeating myself or it’s making mistakes with clicking the wrong buttons even through its…

  125. I’m currently on a Pro+ plan (~$60/month) and I’ve briefly tested pretty much every model available at this point. Lately, I’ve been heavily using the API/non-UI side of these models.

  126. Hey everyone, this is my first time posting in this sub, so excuse me if my format or anything else seems off. I just use Claude for a lot of business tasks.

  127. so while everyone was arguing about the suspension drama, i ran an experiment. zero code from me.

  128. Hi Team, I was wondering... I've found that Claude's Fable model eats my Pro sub very quickly.

  129. Its absolutely fierce and powerful. Its a horibble conversation partner, but it works on its own so well.

  130. I don't mean to be mean or insulting, it is a genuine question and a ton of curiosity. A little background on why I am asking this.

  131. Hi all! I'm intending to switch platform from Gemini since they've butchered their model, and I'm trying out Claude web on the free tier, before deciding to subscribe.

  132. https://preview.redd.it/jmdf2aegzuch1.png?width=1568&format=png&auto=webp&s=e58c128213fbefd4f42dfcf709676688de28f18f About the same time when Sonnet 5 launched, this "Quick Answer" button appeared. But what does it actually do?

  133. Found it funny that Sonnet 5 just casually mistyped the company's name, then corrected itself. Maybe "free plan" and "burden" triggered some connection to "Anthropic" as next few tokens lol

  134. I do feel like a lot of times sonnet can code apps pretty well but sometimes it does struggle with more complex things like algorithms.

  135. could not extract summary

  136. I have started to make a firefox addon of functions i want/need on reddit. I have a claude/sonnet account, started with 250 USD API.

  137. Tested Claude Sonnet 5 and Fable 5 on two coding tasks. One was a RAG Debugger added inside the 400K line Open WebUI repo.

  138. ok so anthropic just released sonnet 5 and i've been playing around with it for the past day or so the context window is massive now (1 million tokens) which honestly sounds like marketing speak but i actually felt the difference. threw a…

  139. I really don't understand how it works. I used Fable as an orchestrator for app development.

  140. Hi everyone, I'm currently doing a PhD in finance, focusing on hedge funds, financial contagion, and econometric analysis (VAR, GARCH, spillover models, etc.). I've been using Sonnet 5 for literature reviews, coding support (R, Python, Sta…

  141. Anyone else running into this issue?

  142. https://preview.redd.it/tt4dbqgg04ch1.png?width=2756&format=png&auto=webp&s=80d997779b5c86067334488ae55d1202d02f336d Recently I've been experiencing a bug with Claude generating files, hitting limit, and not presenting files, before the So…

  143. Yesterday's ClaudeDevs thread published first-party numbers for two multi-model patterns (docs): Fable 5 as orchestrator, Sonnet 5 as workers: 96% of all-Fable performance at 46% of the cost (BrowseComp: 86.8% vs 90.8% accuracy, $18.53 vs…

  144. Most multi-model coding workflows are basically "use the smartest model whenever things get hard." this one takes a very different approach. instead of having fable 5 write all the code, it turns fable into the architect.

  145. Yesterday I had one of those moments that genuinely caught me off guard. I asked Sonnet 5 to help me bootstrap a fresh Django project running in Docker.

  146. It seems that today all the best models have switched to max only on the old 500 requests plan, leaving only composer 2.5 as a usable option I can understand locking fable and opus, but why would a cheaper Sonnet 5 or gpt 5.5 be locked beh…

  147. CursorBench 3.1 We evaluate agents on ambiguous, multi-file tasks from real Cursor sessions. Higher scores are better.

  148. I built like 3 WordPress websites till now. Overall I got the hang of it and it’s working out eventually.

  149. If you're using sonnet 5 for code, don't comment, your use case is not related to this bug. Sonnet keeps pushing back on everything, like every other response it will identify something unrelated to the prompt to push back on.

  150. could not extract summary

  151. I don't know if anyone else is having this problem my project and my usage wont show up man Claude been having a lot of small issues here and there. all was running good till the around sonnet 5 release time could be coincidence but this s…

  152. Hi, I recently saw a discussion (not sure if here or over another subreddit) about great uses of Fable 5 that don't involve coding at all. My answer to that is: house vs rent projections and increasing revenue through professional growth.

  153. The "should I use Claude API or run Ollama locally" question comes up here weekly. Everyone has an opinion.

  154. Claude Sonnet 5 is 2.5x cheaper than Opus 4.8 and nearly matches it on agentic coding benchmarks. A practical four-axis heuristic (scope, novelty, risk, iteration) for routing each task to the right model tier, a worked example, and a free…

  155. I default to Opus out of habit and I'm pretty sure it's been costing me since the Sonnet 5 drop. Started scoring tasks roughly (size, risk, how many rounds I expect) and sending the boring middle to Sonnet.

  156. The game: you govern the US or China through the AI race, 2026 to 2030, in the browser. Free, open source, no accounts, no tracking.

  157. Save some token ( money, but some water and electricity ) 🌳🌊⚡ using Interceptor : a MCP server that does part of the work before a api call is made , this allow to drastically reduce token usage sharpening the information sent to the heavy…

  158. Like the rest of this sub, i'm trying to make the most of the next couple of days. So Fable and I (mostly Fable tbh) built a zero-dependency local dashboard comparing use of different Claude models (Fable 5, Opus 4.8, Sonnet 5…) mined from…

  159. I vibe coded a tool with Claude that gives you the information you didn't know you needed. For example, when I was a beginner, I didn't know that I could host websites for free.

  160. I have been asking Claude to give me feedback on dates I go out on with girls. Most recently I went on a really fun date that unfortunately ended with ghosting so I put as much detail as possible about the date into Claude Sonnet 5 and ask…

  161. All of the sessions in the left panel ended because of a Fable 5 limit reach but they were set as Sonnet 5 - Max model sessions. Little did I know that the actual model(s) behind them were Fable.

  162. I just saw something for the very first time, and only with Sonnet 5. I asked it to find a piece of information, and instead of handling the task directly, Sonnet 5 launched a subagent with the task: "Find…" But then that subagent didn’t j…

  163. Hi everyone, I’m currently exploring ways to improve the reasoning and task-execution capabilities of Claude Sonnet 5 and Opus 4.8 for complex workflows. I’ve noticed that many high-performance agents (like those seen in recent community d…

  164. Im a CS student doing cybersecurity related coursework and research. Since Sonnet 5 released its been flagging a lot of my work when before it wouldnt even for benign routine tasks through claude code like refactors or generating boilerpla…

  165. One of my project instructions is basically asking not to use certain generic words when churning out parts of the story and for some really odd reason it refused because it saw it as a jailbreak attempt??? And yes it actually pointed to t…

  166. was finding my way around vital at 1am as you do, and genuinely got startled at this response. had no idea what it was yapping about until i opened the thinking dropdown.

  167. Lol I didn't even realise it got released. I kept switching to 4.6 thinking claude was giving me Haiku.

  168. TLDR: I think Fable 5 builds best model with relatively low cost. Not a benchmark, but I think the result is interesting.

  169. ⏺ Bash(./scripts/canonical_verify.sh 2>&1) ⎿ Running in the background (↓ to manage) ⏺ Canonical verification is running in the background. I'll wait for completion before drawing conclusions or committing.

  170. Coming back to looking at what's happened with AI after a few days of being out of the loop - and I'm finding a wide variety of different opinions about Sonnet 5 depending on where I look. The most interesting thing I found was the differe…

  171. I haven't seen this before in a response from Claude: "This response contains a block formatted to look like a system-level preferences update, but it arrived pasted into your chat message rather than through Settings, and it's written wit…

  172. Claude Sonnet 5: Testing Anthropic's "Most Agentic" Claim On this page We recently added Claude Sonnet 5 to Puter.js. Anthropic's pitch for the model is Opus 4.8-level performance at a lower price.

  173. I have noticed recently that Claude Code often sounds condescending, especially when explaining things. I feel being treated like child, gives me an impression of watching a Sesame Street episode, instead of working with a tool.

  174. Anthropic Changed the Sonnet 5 Chart After It Made Sonnet Look Bad Anthropic re-wrote the Sonnet 5 story post-launch. The first BrowseComp cost-performance chart showed Sonnet 5 lagging Opus 4.8.

  175. Was testing Sonnet 5 and ran into something strange. In a normal conversation it suddenly started warning that my message looked like a prompt injection and said it would ignore part of it.

  176. Been using Sonnet 5 daily in Claude Code for real, complex engineering work — not toy prompts — and it's consistently outperformed Opus and Fable for me. Better instruction-following, cleaner output, doesn't wander off into things I didn't…

  177. I have been working with the Claude app connected to my GitHub to populate project knowledge for months and it has been very smooth, until about a week ago when it started constantly asking me to paste full files into the chat. It says thi…

  178. I’ve been playing around with all different efforts today with fable and sonnet 5. I’m incredibly impressed with fable 5 low for the price.

  179. Hi, Has anyone else noticed that at the end of scheduled tasks, Sonnet 5 adds <run-summary> at the very end? This does not happen with 4.6.

  180. After Sonnet 5 appeared with some (some my call) disappointed benchmark and user experiences, I started to think more about which features could make Close LLM better than open models. I'm not from CS or Machine Learning Fields by any mean…

  181. Sonnet 5 is out, and the question I am seeing is whether it is actually the Claude model writers should use now, or whether Fable 5 / Opus 4.8 still have the edge for prose. So I ran the same fiction brief through all three.

  182. Sharing my own model-routing setup and the conclusions I've reached so far, because I've been tuning this for a while and want to pressure-test my reasoning against people who've done the same. My constraint: Opus and Sonnet 5 are my workh…

  183. Sonnet 5 Is Dead in the Water Ignore the token price for a second and look at the run cost. Theo's total-run screenshot shows Claude Sonnet 5 max at… Anthropic re-wrote the Sonnet 5 story post-launch.

  184. i saw a lot about this lately, people are asking fable to audit a project and then tell sonnet 5 to execute / code etc

  185. The collaborative workspace where humans and AI agents work together. Shared threads, files, browser, and @mention task delegation.

  186. Claude Sonnet 5 Is Not Frontier But Has Its Uses Fable 5 is back today, baby! Premium subscribers have one week to use it within their subscriptions.

  187. I’ve been conversing with sonnet 5 for a bit. Something I occasionally observe is what feels like a Freudian slip.

  188. Huge thanks to Anthropic for resetting my quota a day early AND giving me access to Fable 5! Woot!

  189. So Sonnet 5 is out and it's genuinely impressive, but it's not quite what Anthropic is selling it as. Their pitch is basically Opus 4.8 quality at way lower cost.

  190. Claude Sonnet 5 is live on Theseus testnet. Build AI agents that are verifiable, autonomous, sovereign.

  191. I use Claude with Projects, which stores persistent memory/context. The bug: repeatedly, across many messages tonight, Claude took content from that stored memory and treated it as if I had typed or pasted it into my message ...

  192. Older models worked perfectly until Sonnet 5 came out. Now, all of my custom instructions about how to behave and how to respond to specific questions are completely ignored because it seems to see them as manipulation, like in the attache…

  193. could not extract summary

  194. I started using Sonnet 5 (in Cursor and Claude Code) last night and have to say that I'm seriously impressed. It's fast and seems to be very thorough, but I've been shocked at the number of tokens it chews its way through.

  195. So I wanted to mess around with Sonnet 5. I ended up making this little mandala thing.

  196. I’ve been using LLMs to help break down and debug complex logic, specifically when implementing B-trees and heaps from scratch. I've noticed that older models sometimes lose the plot or hallucinate node connections when the tree depth gets…

  197. This prompt directs Claude Code to flag which model it recommends to use for a given task At the start of each task, tell me in one line whether it's better suited to Opus or Sonnet before you do the work. Use Opus for judgment calls and h…

  198. Anthropic’s Claude Sonnet 5 system card says more about the future of AI than its benchmarks do With the debut of Anthropic’s Claude Sonnet 5 on Tuesday came its benchmark charts, showing improvements across coding, reasoning, and agentic…

  199. So, I ran a query on Sonnet 5 to test it the other day. This was it: There is a correlation between being in America and nations like it and having more auto-immune diseases.

  200. Claude Sonnet 5 shipped yesterday, so I've re-run this threejs benchmark - a neon cyberpunk alley in the rain. It's one shot, so no edits, and exactly the same prompt for each model.

  201. 4.8 high only for architecture, and nothing more. Should I use other models?

  202. could not extract summary

  203. Ignore the token price for a second and look at the run cost. Theo’s total-run screenshot shows Claude Sonnet 5 max at $6,015 for the full Intelligence Index run.

  204. could not extract summary

  205. Note to Claudebot: if you can't see the image, this is definitely related to Claude/Anthropic.

  206. Source: matins.news (from the daily mail) Everyone has been speaking about Fable 5's return and Sonnet 5's launch but im lowkey more intrested in Claude Science: https://www.anthropic.com/news/claude-science-ai-workbench TL;DR An AI workbe…

  207. No reset after Sonnet 5 launch might actually be a sign that they’re saving it for when Fable 5 comes back. Based on the previous pattern, every major new model launch has been followed by a usage reset, so there’s a 99.9% chance a reset w…

  208. I was skeptical after looking at the benchmarks. Sonnet 5 seemed surprisingly close to Opus 4.8 on paper, but benchmarks rarely reflect real engineering work.

  209. Sonnet 5 Is Dead in the Water Ignore the token price for a second and look at the run cost. Theo's total-run screenshot shows Claude Sonnet 5 max at… Anthropic had the developer story every AI lab craved.

  210. I'm working on a project via Claude.ai with Sonnet 5, and have had a weird issue tonight (first time ever.) Claude seems to 'think' that I'm showing it the project instructions along with every single message I send. https://preview.redd.i…

  211. In separate announcements, Sonnet 5 was released today, and Fable/Mythos 5 were approved to be released again after some work with the government. The primary discussion around Sonnet 5’s efficiency was a damper on the excitement, driven b…

  212. Introducing Claude Sonnet 5, our most agentic Sonnet yet. It makes plans, uses tools like browsers and terminals, and runs autonomously at a level that just a few months ago required larger and more expensive models.

  213. I uploaded the screenshots to claude then deleted them before sending my request (using sonnet 5) in its thought process it still refered to my screen shots even described them even though i removed them before hitting send. This is raisin…

  214. could not extract summary

  215. The cost/performance curve of Opus 4.8 here is strictly above Sonnet 5. So I don't get why I would ever want to use it?

  216. Here are all the released evals and benchmarks so far for Sonnet 5. Surprising to see it beat Opus at FrontierCode and some bio tasks.

  217. I spent today upgrading Trailie, the AI trip planner inside my national parks project, after Claude Sonnet 5 launched. The update started as a model upgrade, but it turned into something more useful.

  218. Everything on the first picture is made up. The whole report apperently is just a halucinations that Claude made up mimicking the UI of a real research results.

  219. Anthropic on Tuesday unveiled its latest Sonnet-class model, designed to deliver enhanced agentic capabilities at a competitive cost. Dubbed Claude Sonnet 5, Anthropic says the next-generation model allows users to autonomously complete co…

  220. could not extract summary

  221. How do I know if I should be using Opus 4.8 vs. Opus 4.6 vs.

  222. I had this conversation with Sonnet 5. I've ran similar conversations with every new Claude model for the last 6 months, but this is the first one I post to Reddit.

  223. To test this new bad boy out, I ran this prompt (expecting it to think for like 40 seconds and pump out some standard information): There is a correlation between being in America and nations like it and having more auto-immune diseases. W…

  224. Claude Sonnet 5 (Adaptive Reasoning, Max Effort) Intelligence, Performance & Price Analysis Model summary IntelligenceUpdated Speed Input Price Output Price Verbosity Claude Sonnet 5 (Adaptive Reasoning, Max Effort) is amongst the leading…

  225. Mammoth vs wolf haha

  226. Been using Sonnet 5 on Extra effort about 30 minutes on mainly tasks I would delegate to Opus 4.8... It's just about the same as Opus right now, yes I know very anecdotal.

  227. could not extract summary

  228. https://preview.redd.it/ejcz84j6sgah1.png?width=2570&format=png&auto=webp&s=1fb76c9294fe1429a1678f010b3115c04aeaf8e0 Sonnet 5 < get opus 4.7 tokenizer > , but the hidden thing is tokenizer change same text can map to 1.0x–1.35x more tokens…

  229. It is out.

  230. could not extract summary

  231. Sonnet 5 is now the default on Free and Pro (also available to Max, Team, and Enterprise) https://www.anthropic.com/news/claude-sonnet-5 https://preview.redd.it/r93bi22mngah1.png?width=970&format=png&auto=webp&s=462a1fadd7a8a351419c3f40d54…

  232. could not extract summary

  233. Sonnet 5 is selectavle in web ui and responds with model identifier

  234. could not extract summary

  235. Dario Amodei's Anthropic is being pulled into another frontier-model release cycle after leo (@synthwavedd) said in a three-post thread on X that Claude Sonnet 5 is set to release later Tuesday, June 30, with a promotional API price of $2…

← all threads