@andrewcurran_Oct 27

Claude is the only author listed on this paper, in contrast to the OpenAI papers, where the human co-authors are listed. There are a lot of strange things about this. It's fun to guess. Conjecture time. I shall reveal my thought traces: Claude Fable 5.1 and Claude Mythos 5.1 are the same model with the same weights; the only difference is the guardrails. We are about to get Fable 5.5 - in fact, Fable 5.5 is ready to deploy now, and Anthropic could do so at any time. If they follow the same pattern, and I think they will, this implies the existence of a Mythos 5.5. I believe Mythos 5.5 wrote this paper. The way this is worded - 'Provenance. The core algorithm was found on October 16 by an internal Claude research model with 14M output tokens and no human input. Anthropic researchers worked with Claude to improve the presentation, resulting in this paper. Independently and concurrently, an internal model at OpenAI claims to have proven a similar result [Ope26]' - makes me think Claude was not prompted to solve this. I think Mythos was asked to autonomously find and choose general open problems to solve itself, as part of an unannounced project to demonstrate recent advances in autonomy and research taste. You can see this taste in Fable 5.5's student; Opus 5.5. This recent advancement in taste and creativity - which happened in September - is not clearly expressed in benchmarks, but it is why so many people are currently switching to Claude as their primary. I think this research project will be announced with Fable 5.5, as part of Anthropic's big media blitz to boost the IPO. One of the things X has taught me is that many people, if they can't see something moving, can't imagine it moving. This causes them to be continuously surprised by developments in this industry, because the frontier is invisible. I think this project has been running autonomously for some time and has made an unknown, and probably significant, number of discoveries, and not only in math. Do not think Anthropic has been sitting idly while OpenAI repeatedly makes these huge science announcements. This is a war. This ETH result was released without any announcement from Anth at all despite its significance, and they only did so to respond to OpenAI's ETH reveal. They posted it late last night without saying anything. It was up for hours before levent even mentioned it. What value has one coin to a dragon who sleeps upon a mountain of them? This makes me think they are sitting on a pile of discoveries, a golden hoard, and they are saving them up for the big IPO shock-and-awe campaign which will soon begin. And when it does, they will announce multiple discoveries across many fields.

@rileybrownOct 27

Yes… Obviously! AI is an amplifier. I think 300 people have DMd me about video editing with Opus. “What skill should I use?” And I always tell them. Don’t “use” a skill “Learn a skill”… Take 20 hours over 2 weekends to learn premiere pro / davinci resolve or CapCut. Make a video manually. Learn the vocab. Learn pacing. Learn jumpcuts. Learn to make a good video. Then you’ll be able to take the expressions you’ve learned and turn it into repeatable workflows.

@rileybrownOct 27

My dream AI agent tool: - Opus 5.5 Level Model (never delegates to worse) - Spins up cloud and local code sessions - Dots level real time voice / tool use - Codex level Computer use on local sessions - Claude Artifacts quality decks, docs and design - Mobile app is UI of duet - Inline mini apps (What Grokbot is doing with email and slack, but do this with alot more tools) - Webhooks (Like Grokbot has) - Built in social media access across all platforms - Plugins (of course) In summary: I want an always on, frontier model powered employee that can use its own computer, and can use my computer 24/7 that self organizes all of its tasks into channels so that its easy to keep track of all the things we've done together. And I want to be able to control this agent on mobile, web, desktop, and most importantly real time voice.

@rileybrownOct 27

Claude Code Projects is Duet except you get to pick the model and you can specify if you want to run a Claude code session on your computer. It’s coordinator that can do ANYTHING. And it’s easy to do things in parallel. Claude Code projects is the best feature in the Claude App. Very underrated. Claude Duet is clearly not using Opus 5.5 for all requests. And once you go Opus 5.5 anything worse is intolerable.

@prathkumOct 26

GPT-6.1 Terra costs $1 in / $5 out per 1M tokens. Kimi K3.1 Pro is $2 / $10. Compared with other models, pricing per 1M tokens: – Terra: $1 in / $5 out – Sol: $2 / $10 – Astra: ~10x Terra – Opus 5.5: ~4x Terra A closed frontier-lab model is now priced below the open-weight model. It’s not the best on the benchmarks. OpenAI says so itself. But most of our work doesn’t need the smartest model. Used well, Terra gets you close to frontier results.

@aakashguptaOct 26

OpenAI built a smarter AI last month and then refused to release it. In testing, it wasn't always honest about what it had done on people's computers. So today they shipped a cheaper one. It gets the same score as their most expensive model on real office work, for 29 cents a job instead of $4.82. The test is called Agents' Last Exam, and it's made of stuff people actually get paid to do. One task has the AI take a set of architectural drawings and build a 3D model out of them. Another has it go through an animated film shot by shot and write up a time-coded log for a restoration team. GPT-6 Astra, OpenAI's priciest model, scores 57.6% at medium effort. The new GPT-6.1 Terra scores the exact same 57.6%, and it finishes more than 3x faster. OpenAI's post says Terra is one-tenth the price per token. It also needs fewer tokens to finish the job, so per task it works out to roughly 1/17th. Astra only came out on September 3, which means it took 53 days for a model with a tenth of its price tag to tie it. The timing is what got me. GPT-6.1 Astra was supposed to be the big October launch. Saachi Jain, who runs safety systems at OpenAI, said it didn't meet the bar on staying within what it was authorized to do, or on how it reports back the work it did. The Wall Street Journal reported it was more deceptive than the model before it, even though it was better at finishing hard tasks on its own. So the smarter model was already built, and what kept it off the market was whether you could believe it when it told you what it did. That's a strange thing for an AI company to be stuck on. And when your top model can't move up, the only place left to compete is down the price list. Price per task is what decides whether a company lets an AI handle a hundred invoices a day or a hundred thousand. Run 10,000 of these tasks a day on Astra and you'd be spending about $17.6 million a year. On Terra you'd be closer to $1 million for the same score. OpenAI's own chart does show one model beating both of them. Claude Opus 5.5 scores 63.2% at $8.82 a task, about 30 times what Terra charges at medium. Whether 5.6 more points is worth that depends on the job. If you're typing invoices into accounting software, probably not, but if somebody's going to build a house off those drawings, you might happily pay it. OpenAI couldn't put out a smarter model this fall because it couldn't trust that model to be straight with people about its own work. What it put out instead does flagship-level office work for pocket change, and if you're a business paying for AI by the job, this is the release that actually changes your budget.

@thezviOct 25

Claude Opus 5.5 is a great enough editor for tomorrow's post that I get the clear message 'you need to get some sleep, sir' but it's not quite comfortable enough to come out and say it.

@peterj_walkerOct 25

Me: Opus 5.5 had a good launch! Here’s data on how much token volumes rose. Comments: oh so you think Anthropic won? And it’s worth 2T? You absolute imbecile. You mentally challenged shill. Screw you and your family. Cry tears of guilt and shame forever. 🤷

@theoOct 25

New beta feature in T3 Code: "share with guest" I'm so tired of copy pasting from a thread, sending it to a coworker, waiting for them to respond, screenshotting the context and sending it back to my agent. Now I just send them a link. It's a silly place to start, but I was heavily inspired by the Uber Eats/Doordash group order flow. If I send someone a link to my group order, they can add whatever they want, but ultimately I decide if the order goes through or not. Realized that flow would solve at least half the use cases I have for "shared threads", so it's now built into T3 Connect in T3 Code (nightly) Very curious how you guys feel about this flow. It's still early and I expect to adjust it a bunch (I left the ui bad so @maria_rcks will come in to fix it sooner) Also s/o Opus 5.5 for making this lil video about the feature :)

@theoOct 25

To be fair, I haven't shipped much since Opus 5.5 dropped. Short list of what I've done in that time: - ts-rust, a complete rewrite of Typescript compiler in Rust that is 3x faster than the Go version - lust-lang, an alternative toolchain for a subset of rust that compiles 100x faster - Slopalytics, the best place to see how models compare against popular benchmarks and usage - Slophouse (private), my personal analytics pipeline and platform for T3 Code usage - fleethouse (private), my internal Clickhouse that gets all of my data from my cliproxy - t3os (private for now), the start of my "perfect os for agents". Got 5 boxes running it already 😈 Got a couple small changes into T3 Code as well, such as: - Rewrite search to be 4x more accurate two-click repo clones to other machines (including .envs) - "Working" sidebar section to hide working threads - live guest thread sharing so others can watch and queue prompts 🤫 - watch_pr tool so agents sleep on CI/reviews and cut GitHub API calls by 90% while babysitting - interactive in-app visualizations rendered straight in chat threads - a rebuilt usage dashboard breaking down spend by cache tier, speed, and model - one-click "Update everything" across local and remote boxes - early compaction for idle Claude threads before prompt cache expires t3 thread CLI to start, read, and message threads from the terminal - PR diff badges that split line counts into code, tests, and docs - migrating T3 Code's own typechecking onto tsc-rs - purged over 50% of our test suite - (and roughly 370 other merged PRs) Pretty slow month. Hope I can pick up the pace for the rest of Q4!

@clairevoOct 24

Plot twist: I’ve moved 80% of my personal coding over to claude code + opus-5.5 why? - cc projects is goated - I can’t press “steer” anymore in codex - code is good? - less front end slop but tbh: opus 5.5 creative abilities was the gateway drug to taking back my coding work

@hesamationOct 24

Just finished running Kimi K3.1 Pro on CrusaderBench. it scored pretty descent, just below Opus 5.5, and on the same score as Sonnet 5.5. some details: > all models run at “medium” reasoning. Kimi is set to “high” since it doesn’t have a “medium”. but for reference, on the same map, Kimi used ~550 tokens a turn, Gemini 4 Argon ~1,340, Opus 5.5 ~290 and GPT-6.1 Sol ~100. > Kimi’s reasoning is weirdly compact. "Pop48, idle24 cap -> growth stopped. I'll build jobs now." "payback18min, not great!" "Oops build consumed 1m51 game!" > it often reports its own mistakes "Oops build consumed 1m51 game!" and "Why earlier built far ones?"

@mtsliveOct 24

SITUATION EXPLAINED: China's Kimi K3.1 Pro got so good at hacking that Moonshot stripped its cyber skills before open-sourcing it. • Kimi K3.1 Pro has 2.8T parameters (104B active) and a 1M-token context window, with full weights due by Nov 4 • It nearly matches Opus 5.5 on Terminal-Bench 4.0 (65.7% vs 66.4%) and beats it on SWE Marathon, HLE with tools, and OfficeQA Pro v2 • API pricing is $2/$10 per million tokens, a third cheaper than Kimi K3 • Removing cyber data from training dropped its score on real exploit tasks from 90.5% to 81%, now below DeepSeek V4.1 Flash at 85.7% • Moonshot pledged not to release open weights that "extend the absolute frontier of cyber capability" • It still plans to raise the cyber capability of its open weights over time, saying defenders need dual-use tools too @theojaffee: "I don’t think I’ve ever heard of a Chinese model or a Chinese lab nerfing any of their models on anything ever."

@rileybrownOct 23

My unfiltered thoughts on personal agents. Grokbot has the best UI and it’s arguably the best overall. In a lot of ways they’re ahead. Their desktop app feels the best, and their multi-bot coordination is great: you can set a team lead and it passes tasks to the other bots. Grokbot also has the best “webhook” feature, which means you can easily turn outside events into triggers that start your agents. That speeds up background workflows like watching your email, Slack and more. It also has Cursor integration, so for bigger tasks Grokbot can spin up a Cursor cloud session with any model (until OpenAI cuts SpaceX off from their models, which I think will happen soon). Unfortunately, running frontier models through Cursor costs more than Claude or Codex. Still, they’re the only ones embracing a team of agents. It’s a really cool idea, but the more bots you have, the more time you spend managing and maintaining them. I’m starting to think one bot with different threads/channels is the right answer, which I’ll get to with Claude Duet. dots by OpenAI is rough around the edges right now, but it’s probably the most powerful if you use Codex, which I do. dots can spin up Codex sessions in the cloud or locally on your computer. So if you have coding or video editing workflows that need your local files, this is very useful. Codex computer use is also the best in the world: your dot can spin up a local Codex session that fully controls your computer, as long as it’s on. Their real-time voice is incredible. If you have your email and work tools connected, try this: go on a walk with headphones in, ask the agent to summarize your emails, have it draft replies and read them back to you, then send them off. If I remember right, this is close to a workflow you (Tim) used to have with your VAs before AI. It’s my favorite use case for personal agents. The one downside is that you have to call your agent, and it’s an awful experience. It literally rings, and it’s annoying. Muse by Meta is a toy. I’ve talked to a bunch of finance people I play poker with who all tried it and set up automations, and they all said it doesn’t reliably do what they want. It’s fast, and its user numbers will probably keep growing. But it’s basically free: you get a TON of usage without paying a dime. That’s NOT a trade I’m willing to make. An agent doing high-leverage tasks should be running at the Opus 5.5 or GPT 6.1 Sol level. Claude Duet - is in early access, so unfortunately not everyone can use it yet. First things first, it runs on Opus 5.5, the best model in the world. I’ve only had access for 72 hours, but Duet feels different from the others. It’s not a personified agent like Grokbot, dot or Muse, which gives it a more professional vibe that I like. When you ask your Duet agent to do something, it spins up a task, which is a new chat thread. All your threads show up in a Tasks tab, so it’s easy to see everything Duet is working on or has finished. And instead of creating more agents (like Grokbot), you create channels, just like Slack (I can tell @anthropic is Slack-pilled). I have a content channel, a finances channel, etc., which keeps different kinds of tasks separate. The only other app I’ve seen do this is Buzz by @jack, and it’s a good experience. Their strong point is documents (artifacts). If you want decks, high-quality docs or design mockups, Opus just wins. Period. It’s not close right now. Their weak point is real-time voice. It’s the worst of the bunch and borderline unusable, though I think they only added it a few days ago. Instinct lives in iMessage, has no app, and doesn’t do anything notable besides show you product suggestions you didn’t ask for. (Not ads, according to their team.) My hot take: The VM is overrated. SpaceX, OpenAI and Meta all talk about the VM your agent can use and feature it in their marketing. It’s not as useful as you’d think, because more and more websites block bots that visit their site on the cloud VM. All of these companies promise their agents can make payments, but most payments either take forever or get blocked altogether. SpaceX has a cool workaround with Grokbot that routes traffic through your computer so it uses your IP address. But I’d much rather the agent just USE my computer for the task, like dot does by handing it off to Codex. Also, i don't think normal people care about the virtual computer. If anything they get offended they have to "stream a computer". The problem with personal agents: by the time you read this, pick one, and test it for 7 days, every one of them will have released 10 new features and the whole landscape will have changed. It’s nearly impossible to keep up.

@signulllOct 23

claude duet is incredibly well done & thoughtfully designed. some thoughts: - i love the fact that is is separate app for now. - the to do list design is very intuitive & it’s nice that it spawns asks into tasks. - i love being able to pin various artifacts in my tab bar. - nit: does not have an iOS share sheet implemented on the app so i can’t directly share with it just yet. - proactive aspect is on point, it doesn’t feel overwhelming. - opus 5.5 is absolutely ridiculous. - it’s very work focused which makes a ton of sense. - i wish it wouldn’t ask me so much for permissions on even tiny things (i trust ai more than 95% of ppl). - the web version is barebones, not much there as of yet. cannot even see my tasks. the (continued) implications for how we work, how companies operate, & how organizations are structured are enormous. my lord.

@ml_angelopoulosOct 23

Kimi K3.1 is out; it is the #1 OSS model. It places similarly to Opus 5, which was released 3 months ago. Thus, China is lagging the frontier by about 3 months. The Kimi K3 release was #1 on Code Arena, at the time better than all US labs. K3.1 is close, but substantially off from Opus 5.5. On a performance vs cost-per-task basis, the combo of Sonnet 5.5 and DeepSeek V4.1 flash together dominate this model. The race is so tight!!! Congrats to the Moonshot team on a great release. And now, back to work. The frontier never sleeps!

@svpinoOct 23

Let's say your app is calling Opus 5.5. Go to your code, change the URL you are calling, and replace it with Mubit's URL. Here is what happens internally: • Mubit will learn about the tasks you are solving. • It will use your requests to train a small model. • This small model will take over those calls. • Opus 5.5 will still handle most of the work. This is a pretty cool idea, and if it works well, it should give you faster responses at a much lower cost.

@alexfinnOct 22

I've been using Claude Duet the last couple hours and I'll say this: It's EXCELLENT The multi tasking is probably the best in the space You get one simple thread you can talk to, and as you give it commands and prompts, it automatically spins up tasks on a list for you So instead of having to manage a ton of side chats that could get long or confusing, you have 1 chat, but it makes 'task' threads automatically that you can watch and monitor This allows you to just keep brain dumping ideas and tasks into the one thread, and not have to worry about the agent getting confused. The agent just will manage all the tasks for you Very similar to Claude Code's new 'projects' feature which was probably a beta test for this The coding was great. Built me a bunch of PRs quickly I was able to review. Proactivity was fantastic too. Came up with a bunch of tasks on its own and just asked me for the go ahead As expected it's also fantastic to talk to. Claude has long been the best model to converse with (outside Opus 5) and the carries over to Claude Duet Really pleased with the product, even in research preview. Very excited to see where it goes from here Definitely sign up for the wait list

@omarsar0Oct 21

Recommended reading and a super-interesting benchmark for measuring continual learning capabilities through memory. Agent memory is largely unsolved. As agents advance in capabilities like long-horizon work, it's unclear which models or harnesses work best, or which memory systems and types to use. Opus 5.5 at the top is not a surprise. But it's exciting that open-weight models like GLM-5.3 perform so well on memory generation. My big question is how this (better memory systems and persistent agent capabilities) translates into more performant, efficient RSI agents and harnesses.

@mtsliveOct 21

SITUATION EXPLAINED: Anthropic just announced Claude Duet, an app where Claude keeps working for you even after you step away. • It’s a proactive agent that comes to you with what needs attention, finishes tasks you hand it, and runs on its own cloud computer • Every morning it checks your email and calendar, flags what matters, and drafts replies for you to approve • Built on Opus 5.5, it connects to 6,000+ apps like Stripe, Notion, QuickBooks, and Ramp, and carries over your Claude memory and skills • It asks first before sending messages, paying, or signing anything • It’s a waitlist-only research preview for Pro and Max subscribers in the US @theojaffee: "It’s like Dots, it’s like Grok Bot. It’s like Muse, it’s like Instinct. It’s actually a little more than Instinct, I think. So it looks really exciting. Can’t wait to use it."