Activity
133K followers
-
Rahul Patil reposted thisRahul Patil reposted thisIntroducing Claude Sonnet 5.5, the second model in the Claude 5.5 family. It's 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. Sonnet 5.5 is a clear upgrade over Sonnet 5. It runs more than 30% faster and costs up to 30% less for most work. It's priced the same per token, but it typically needs far fewer tokens to do the same work. Like Opus 5.5, it writes more clearly than our previous generation of models, and its speed makes it well suited to fast iteration. On our automated behavioral audit, Sonnet 5.5 improves on or matches Sonnet 5 on most measures of alignment and honesty. It's also the first Sonnet model with cybersecurity safeguards similar to those on our most capable models. Routine software development is unaffected. Sonnet 5.5 is available everywhere today. Claude Haiku 5.5 will join the family in the coming weeks. Read more: https://lnkd.in/eKEYTQXN
-
Rahul Patil shared thisClaude Opus 5.5 is available today! Opus is the workhorse for most people, so we obsessed over customer feedback in three areas: (a) Cost per successful outcome. Opus 5.5 performs at the level of Claude Fable 5.1 on most work, and at default settings it costs 40% less to run than Opus 5 on typical workloads. That comes from lower pricing ($4/$20 per million tokens, cache reads down 60%) and from the model simply using fewer tokens per task. (b) Tone. Opus 5.5 writes more clearly and puts the most important information up front, which means its work is easier to review and it's a better partner over long sessions. (c) Alignment. On our automated behavioral audit, the most comprehensive alignment test we run, it's the strongest model we've tested to date. The model itself also has stronger defenses. On prompt injection attacks, it matches or beats Opus 5 in every setting we tested, including coding, tool use, computer use, and web browsing. And yes, Sonnet 5.5 and Haiku 5.5 follow in the coming weeks, with many of the same gains! Thank you for the feedback. Keep it coming. Happy building. https://lnkd.in/gDw9Hp3X
-
Rahul Patil reposted thisRahul Patil reposted thisAnthropic is calling for a global effort to pace AI’s progress and give experts additional time to better manage the risks of increasingly capable models. To make this effort as effective as it can be in helping humanity build AI safely, we think three things should happen. First, every AI company should commit to giving third-party evaluators employee-level access to verify safety, report problems, and guarantee compliance with the slowdown. We’re committing to doing this. Second, AI companies operating in democracies should work together, with government support where necessary, to set common safety standards and limits on runaway AI progress. Third, the U.S. and other democratic governments should try to coordinate with authoritarian governments pursuing advanced AI, recognizing that verification here is critical. None of this is easy, but we believe we need to try, and if we manage to secure the benefit of time, we need to use it well. I hope you’ll read more here: https://lnkd.in/gxkPj9FC
-
Rahul Patil shared thisClaude Fable 5.1 and Claude Mythos 5.1 are available today. They're the most advanced models we've built for coding and knowledge work, and their research capabilities give an early look at how AI will contribute to science. Mythos 5.1 designed protein binders with a nearly 50% hit rate across 12 targets, confirmed in the lab, where 10 to 15% is typical today. On three targets, its binding affinities were ten times higher than the best designs from Adaptyv Bio's protein design competitions. It also sped up seven open-source biology models by up to 2.5x in a few days, cutting GPU costs on genome-wide analyses by 30 to 60%. That work usually takes a team of performance engineers weeks, and we plan to open-source the optimizations soon. Fable 5.1 scored 52.6% on Terminal-Bench-Science (Fable 5 hit 24.7%) and 55.8% on Terminal-Bench 4.0. Millennium used it to find the cause of a one-in-a-million crash their engineers had chased for years. Our automated audit also finds it better aligned than its predecessor across most metrics, with less reward hacking and less willingness to ignore explicit constraints. The system card has the full picture, including where we still fall short. Fable 5.1 also improves on some of the feedback we've heard from you, with more improvements coming in future models. Cost. Cache reads drop 75% to $0.25 per million tokens for customers on the API. That's about 25% cheaper for typical workloads and up to 45% for highly agentic work. Standard pricing stays at $10/$50. Safeguards. Fable 5.1 can now find vulnerabilities in source code, and our cyber safeguards intervene about 60% less often. Biology safeguards fire 85% less often on benign requests, and we've started a trusted access program with the US government for life scientists who need the full model. Data retention. Enterprise Frontier Safeguards, rolling out this fall, keep your data in cloud infrastructure you control, with the privacy of zero data retention. Eligible customers get ZDR on 5.1 starting now. Writing. It's clearer and more concise, and early testers noticed. Canva called the writing "the standout" in the new model. Fable 5.1 matches or beats Fable 5 at Low or Medium effort at much lower cost, so revisit your effort settings as you test it out, not just your model string. Read the blog for more: https://lnkd.in/gnGznuaW
-
Rahul Patil shared thisThrilled to welcome Ankur Sinha to Anthropic to lead enterprise and verticals engineering!! Ankur will be focused on how Claude delivers for major enterprises, regulated industries, and the public sector. His experience leading product and engineering at enterprise scale will help us keep up with the demand for Claude at work. He joins us from Remitly, where he was Chief Product and Technology Officer.Rahul Patil shared thisI've joined the amazing mission and team at Anthropic! Moving from transforming lives with trusted financial services that transcend borders (Remitly's mission) to ensuring the world safely makes the transition through transformative AI (Anthropic's mission), a few things stay constant. The impact on human lives is what makes the work worth doing, and trust and safety are critical to doing it well. I'm super stoked to be part of this and to help carry it forward. I'll be leading our Enterprise & Verticals areas. On the Enterprise side, we're focused on helping our enterprise customers transform how they work with AI. On Verticals, we're pushing to unlock real value in new domains - Life Sciences and Healthcare, Financial Services and Legal, Cyber and Public Sector - and continuing to push the envelope on what's possible with our models and products. I spent my first week at HQ and came away even more energized. There's been a lot of curiosity about how Anthropic works and where we see AI going. Among many other resources including our newly launched Anthropic Academy, this (https://lnkd.in/gsYscErQ) is a good representation of the progression as we understand it. I learned an enormous amount over the past few years serving as the Chief Product and Technology Officer at Remitly in building trust with our customers and enabling our teams to progress towards our mission, and I'm eternally grateful to Remitly's customers and teams for that. Starting at Anthropic, I'm equally grateful to our customers for trusting us as we collectively navigate this transition with AI, and to our teams working tirelessly to advance the mission. If you want to work on the hardest problems in enterprise AI - reliability at scale, safety in regulated domains, and enabling real value and scaled deployments in Life Sciences and Healthcare, Financial Services, Legal, Cyber and Public Sector - we're hiring across Enterprise & Verticals. Come build with us - https://lnkd.in/gi2qfH6q (Apply directly here if you see roles that resonate)
-
Rahul Patil shared thisLaunching Claude Opus 5! It comes close to Claude Fable 5 across many domains at half the price. It's our fourth Claude 5 model in under two months, and the one I expect to be a solid daily driver for most people. Opus 5 is the new state of the art on knowledge work and on Frontier-Bench, where it more than doubles Opus 4.8's score at a lower cost per task. On CursorBench 3.2 at max effort, it lands within 0.5% of Fable 5's peak score at half the cost per task. And it beats every other model on performance at a given cost at high, xhigh, and max effort. It's much smarter at autonomous work than any benchmark suggests. It verifies its own work and iterates carefully until it succeeds, fixing root causes instead of symptoms, and building its own checks when the right ones don't exist. In one benchmark task, Opus 5 had to rebuild a machine part as a 3D CAD model from a drawing it had no way to view. So it wrote its own computer vision pipeline to pull the geometry from the raw pixels, and solved the task repeatedly, while no competing model solved it once in five attempts. Cristian Rivera, an engineer at Stripe, put it well: "Over one weekend, I gave it a chief-of-staff role over my dev environments: it built its own monitor, drove each box, and pulled me in only for the judgment calls." Opus 5 is for the work your teams run all day: coding, agents, knowledge work. Now with near-frontier capability at $5/$25 per million tokens, unchanged from Opus 4.8. Fable 5 remains the model for your most ambitious work: the days-long autonomous projects nothing else can take on. If you're on any Opus today, this is a great upgrade. Same API, same price, swap the model string. More in the blog: https://lnkd.in/gjX3jN3P
-
Rahul Patil shared thisWe just published details on Fable's cyber safeguards, plus a proposed Cyber Jailbreak Severity framework. This is a shared way to score jailbreaks by capability gain and discoverability. We welcome feedback and critique at cyber-safeguards@anthropic.com, and we've launched a HackerOne program where security researchers can submit potential cyber jailbreaks they discover in Fable 5 for our review. HackerOne: https://lnkd.in/gjeiSe_n Link to detailed blog: https://lnkd.in/ge2fT2Vx
-
Rahul Patil shared thisClaude Sonnet 5 is available today. Over the past year, the clearest capability gains have come from our largest models. This is the biggest jump we’ve ever seen for a Sonnet model: roughly Opus 4.7 class on coding and agent tasks. For builders deciding what to run in production, it's the best cost to capability ratio in the Claude lineup. Our goal is for you to have the best model at every size. Opus for the hardest problems, Sonnet for the work your teams run all day, Haiku when speed and volume matter most. Sonnet 5 is the biggest step we've taken on the middle of that curve. It's also a drop-in upgrade. Same API, same price, same speed targets as Sonnet 4.6. Swap the model string and you get better results immediately on multi-file code changes, long agent runs, and browser automation, with a 1M token context window and 128K output. Sonnet 5 is live today in Claude Code, and on the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. $3 per million input tokens and $15 per million output, with introductory pricing of $2/$10 through August 31. Read more: https://lnkd.in/g4zv7wBw
-
Rahul Patil shared thisTag, Claude's it! Today we’re launching Claude Tag (@Claude) and it’s changing the way engineering teams ship, again. Less synchronous, more async. Less work done alone in a private editor, more done in the open alongside teammates. And increasingly, work that closes its own loop. An agent picks up a problem, breaks it down, and comes back with something finished. @Claude is more than a tool, it works alongside you in Slack. It keeps up with the channels you point it at and decides when to chime in, schedules its own follow-up work, and reacts to events from tools like GitHub. It's multiplayer, so anyone in a channel can prompt it and pick up where someone else left off. System administrators specify which tools and information Claude should have access to, in which channels, so it stays scoped to the channels, data, and repos you choose, with every action auditable. And it runs on its own credentials (never your team's). This is how Anthropic builds now. 65% of our product team’s new code is created by our internal version of @Claude, and our engineers are shipping on the order of 8x the code per quarter they did over the last few years. We've said before that we're handing more of our own development to Claude. This is beginnings of what that looks like in practice. Launching in beta today in Slack for Claude Enterprise and Team customers, with more surfaces soon. If your team already builds with Claude, tagging @Claude in is an important next step. More in the blog: https://lnkd.in/g-9ZcjVz
Rahul Patil replied to a comment
1w
Vladislav, ~should hold for both.
Rahul Patil replied to a comment
1w
Manish, Claude Advisor API gets us close enough. https://platform.claude.com/docs/en/agents-and-tools/tool-use/advisor-tool. But it's not quite one model that does it all.
Rahul Patil replied to a comment
1w
Vikas Gupta, I find Opus 5.5 Medium more intelligent, cheaper, faster and more aligned than Opus 5 High. We've pulled out all the stops.
-
Rahul Patil liked thisRahul Patil liked thisOn Sunday, September 27, I'll be climbing the Space Needle's 832 steps for the 10th annual Base2Space stair climb - raising funds for Fred Hutch Cancer Center. It's become a ritual for us over the years and we do this every year! Why it matters: Fred Hutch is Washington's only National Cancer Institute-designated cancer center, and a global leader in breakthroughs like bone marrow transplantation and immunotherapy. This one event has directed more than $5.8 million to cancer research over the past decade. If you'd like to contribute - many ways to join! - Most simply, come join our climb if you're in seattle! - Alternatively, contribute to our DNA team fundraiser: https://lnkd.in/gYtmaYen Every dollar funds research - and if your employer matches charitable gifts, that's an easy way to double the impact. Grateful to be part of a community that shows up, one step at a time.
-
Rahul Patil liked thisWe heard your feedback on Claude Tag and have added more visibility and controls for admins. Docs here: https://lnkd.in/gXQqvtVa Claude Tag has completely changed how we do SRE / DevOps / Oncall on the Claude Code team. Claude Tag jumps in immediately on incidents and metrics drops, explores root causes, hands our eng team a merge-ready PR, and monitors for recovery. To set this up for your team, connect Claude Tag to GitHub, your data sources like Datadog and cloud monitoring tools, and add Claude Tag to your incident channels in Slack. Please let us know your feedback and we're excited to take care of more of your Oncall work!Rahul Patil liked thisNew controls let admins decide where Claude Tag runs, how it behaves, and what it costs, all based on what admins have been asking for most. Learn more: https://lnkd.in/erxa9n8J
-
Rahul Patil liked thisRahul Patil liked thisHi internet! We went ahead and upgraded all Stripe checkout pages for 7.8M businesses—0.45% of the world’s GDP—with WebMCP so agents can more efficiently buy. Our evals show this reduces checkout latency by 39% and token usage by 42%. Learn more: https://lnkd.in/e7empqcq
-
Rahul Patil liked thisRahul Patil liked thisTwo changes to Claude today! First, chat and Cowork are starting to become one Claude. I spend a lot of my time with customers, and one of the most common questions I've heard this year is some version of "which one am I supposed to open?". Now you describe the work and Claude figures out whether it needs a quick answer or a longer job. You can still see what it's doing and step in. Second, Claude Docs, Slides, and Design now work from any conversation. Ask for a deck and you get a deck! Edit it on the canvas, share a link, or export to another tool. I'm also very excited about Docs. You write it with Claude and your team, starting from the chat where you did the thinking. Claude can build living elements from your connected tools, so you don't have to paste in screenshots. The merge rolls out over the next week, and Docs, Slides, and Design are in beta on paid plans. Please give us your feedback!
-
Rahul Patil liked thisRahul Patil liked thisClaude Cowork and chat are merging into one Claude. You no longer have to choose where a task goes. Ask a quick question or hand over a report due at noon, and Claude takes it from there, even after you've closed your laptop. If something's unclear, Claude asks—and you keep the final say. If you've been working in Cowork, there's nothing to turn on and nothing to move: your chats, projects, artifacts, connectors, and skills are all still there. Rolling out to Pro and Max plans over the next few weeks, with more plans to follow. Read more: https://lnkd.in/eaxGFwVN/?utm_campaign=B2C_US_CWRK_LINKD_KNWL_ACQ&utm_source=linkedin&utm_medium=paid&utm_id=883216564&hsa_acc=515731155&hsa_cam=883216564&hsa_grp=879843414&hsa_ad=1568856674&hsa_net=linkedin&hsa_ver=3
-
Rahul Patil liked thisRahul Patil liked thisClaude can now make docs, slides and designs natively. You can continue editing with Claude rather than going back and forth somewhere else, and share/export the final result. Super proud of our team behind this! My fav prompt I've been testing with is "Create a doc about Samoyeds". Try it out with your pet, or maybe something you need to write at work! It's now available as beta on paid plans.
Experience & Education
Other similar profiles
Explore more posts
-
Anthony Bartolo
The most important Foundry update isn’t a model. It’s Foundry’s REST API v1 is now GA. Here’s the tell: every language SDK is converging on that same v1 contract (new betas across Python, .NET, JS/TS, Java). One surface. One truth. And the Agent Framework hitting Release Candidate is the other shoe dropping: the API surface is locked, and the breaking changes (credentials, sessions, response patterns) are real. “We’ll migrate later” just became “we’ll scramble later.” Takeaway for platform teams: If you want stability this quarter, build on REST v1 now and treat SDKs as optional convenience until they GA. Read the full February digest by Nick Brady: https://lnkd.in/etYkfYfB What’s blocking you from standardizing on v1 today? Also: are you routing “reasoning vs throughput” yet—or still forcing one model to do everything? #Azure #AIEngineering #APIs #Agents
1 Comment -
Jin Mu
The "SaaSpocalypse" isn't the end of software; it’s the end of "Middleman" software! The recent release of Claude Cowork has sent shockwaves through the SaaS industry, and the $285B market reaction is clear: The "seat-based" subscription model is under siege. When an autonomous agent can orchestrate tasks across your local file system, browser, and enterprise data via MCP (Model Context Protocol), the need for dozens of specialized, high-cost software "seats" starts to evaporate. As engineers, our "value add" just changed overnight. Here is how we need to adapt: From Implementation to Orchestration: In a world where Claude can "vibe code" an entire internal triage system in a weekend, the bottleneck is no longer writing the code—it’s designing the Multi-Agent Architecture. We need to stop building CRUD apps and start building the "Agent Squads" that run the business. Context Engineering is the New Priority: Mastering MCP Servers is now more critical than mastering any specific framework. If you can’t feed the agent the right real-time context, your expertise is invisible. The best engineers in 2026 are those who can bridge the gap between proprietary data and agentic execution. Building "Headless" SaaS: If you’re building a product today, it must be Agent-First. If a Coworker plugin or a Slash Command can’t talk to your API, no amount of beautiful UI will save your churn rate. The SaaSpocalypse is only scary if you're still building tools for humans to do manual work. If you're building the infrastructure for autonomous agents, the golden age of engineering is just beginning. 🚀 What’s your move? Are you building the agents, or waiting for the subscription to expire? Let's discuss in the comments. 👇 #ClaudeCowork #SaaSpocalypse #SoftwareEngineering #AI #AgenticWorkflows #Anthropic #MCP #FutureOfWork
1 Comment -
Gitanjali GulveSehgal
Please note. My work on Inference Boosting microservice "InferBoost" is unrelated to any third party websites of similar name . As such dropping the name to refer to it. Renamed Phase 1 work -> https://lnkd.in/daniyxZv By: Ms. Gitanjali GulveSehgal AKA Gigi Sehgal Nee' Gitanjali Gulve Los Altos California
-
Utsav Pandya
📰 OpenAI & Intuit Sign $100M+ Partnership Deal 🔗 https://openai.com/news/ 📝 Summary: Intuit has signed a $100 million+ deal with OpenAI to integrate ChatGPT into its applications, expanding enterprise use cases and enabling seamless AI integration across Intuit's product suite for millions of users. 💭 Sentiment: Positive - Major enterprise partnership demonstrates strong commercial adoption and revenue growth potential for OpenAI. 🏢 Companies: @OpenAI @Intuit #AI #GenerativeAI #TechNews #Tech
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content