The real AI arms race right now isn't about who has the biggest model. It's about who can afford to run it.
Anthropic just showed us the playbook: use your frontier model as a brain, not a workhorse. Let cheaper models do the heavy lifting. 96% performance at half the cost is not a compromise, it's just good engineering. Google's Gemma 4 is pushing the same direction from below.
Powerful AI is getting commoditized fast. The moat is efficiency.
|
|
In Partnership with
|
·
|
|
|
|
|
|
AlphaSignal Workshop: Your LLM's Web Search Is Expensive and Slow. Here's How to Test It.
|
Does the agent you shipped to customers actually have working web access? Or did you bolt an LLM onto a search API and call it done? Most teams never check. Here's the math that should worry you: an org of 1,500 people, half hitting web search 10x a day, burns through 5.5 million queries a year - on infrastructure that was never built for how agents search. On July 28 at 10a PT, Antonio Mallia (founder of Seltz, built web search for Alexa, ex-Pinecone, ex-Amazon AGI) is running a live session that shows you the gap in real time. What you'll walk away with: - What's actually happening under your LLM's web search call ( SERP wrapper, scraper, or real index) and why it changes everything downstream
- Why it's costing you tokens and accuracy, with real benchmark data instead of vendor claims
- A live side-by-side test: same queries, naive setup vs. a real index
- A self-test framework you can run on your own stack that same day
This workshop is for AI/ML engineers and technical PMs building or evaluating the retrieval layer of an AI product, especially if you're running agentic workflows at real volume. Save Your Seat Today - Tix: $50, Location: Zoom, Date: July 28 @ 10a PT
|
|
|
|
Top News
|
|
|
|
Anthropic's Claude agent trick hits 96% top model performance at half the cost
|
|
19,437 Likes
|
|
Fable 5 is Anthropic's most powerful model, but it's also the priciest. Running everything through it is a fast way to burn your budget. So Anthropic shared two smart patterns to get near-Fable quality at a fraction of the cost. Pattern 1: Fable as Advisor. Sonnet 5 does all the heavy lifting. It only calls Fable 5 once per task when it needs guidance. Result? 92% of Fable's performance at 63% of the cost on coding benchmarks. Pattern 2: Fable as Orchestrator. Fable plans the task and breaks it into pieces. Sonnet 5 workers execute each piece. Most tokens are billed at the cheaper Sonnet rate. On research tasks, this hit 96% of Fable's performance at 46% of the cost. Both patterns are supported natively in Claude Managed Agents. Each sub-agent keeps its own cache, so repeated context isn't paid for twice. You can set this up by pinning models in subagent definitions using the model field in Claude Code. This opens up running complex, multi-step AI pipelines at production scale without the frontier model bill.
|
|
|
|
Presented by WorkOS
|
|
Agents need auth. There's now a spec for it.
|
|
When an AI agent tries to complete a task that needs a new account, it hits a wall: the sign-up form. There's no standard for an agent to register a user on their behalf. auth.md is a file you host at your domain that tells agents how to register users, which flows you support, what scopes you expose, and how credentials get issued. Think robots.txt, but for agent registration. It composes existing OAuth standards. Cloudflare, Firecrawl, and Resend already use it. Open protocol. WorkOS-authored.
|
|
|
|
Top News
|
|
|
|
Anthropic brings Claude background tasks to mobile so work continues without you
|
|
16,654 Likes
|
|
Claude Cowork just went cross-device. It used to live only on your desktop. Now it runs on web and mobile too, meaning tasks you kick off at your desk keep running after you close your laptop. Here is what actually changes for you: - Tasks run without you. Close the laptop, Claude keeps going. No device needs to stay online.
- Scheduled tasks work overnight. Set a 6am client brief, and Claude pulls email threads, builds the doc, and leaves a draft ready for your review.
- When Claude hits a decision only you can make, it pings your phone instead of stopping.
- Chat and Cowork now share one tab on web and desktop, so your projects and files live in one place.
- Doubled usage limits are extended through August 5, so you can throw bigger tasks at it.
Beta rolls out to Max plan first. To try it, just start a task the same way you start a conversation.
|
|
|
|
Presented by Reducto
|
|
Reducto ranks #1 in long doc extraction benchmark
|
|
Reducto Deep Extract turns long, dense PDFs into structured data via an API developers can build on. In micro1's LongExtractBench, Reducto ranked #1 with 99.6% precision, 99.6% recall, 99.3% leaf accuracy, and 0 failures. Send a schema, get clean JSON back from invoices, claims, and financial docs.
|
|
|
|
Top News
|
|
|
|
Anthropic offers open-source maintainers 6 months of free Claude Max
|
|
10,857 Likes
|
|
If you maintain an open source project, Anthropic just handed you something real. Claude for Open Source gives qualifying maintainers six months of free Claude Max 20x, which normally runs $200/month ($1,200 total value). So what is Max 20x exactly? Think of it as 20 times more usage than the standard $20 Pro plan. You stop hitting that wall mid-afternoon where Claude says "slow down." You get priority access during peak hours, full Claude Code access in your terminal, and roughly 900 messages per session window. What does this actually unlock for you? - Run full agentic refactoring loops across large repos without getting cut off
- Review and triage issues at scale using Extended Thinking mode
- Keep Claude Code running all day without rationing prompts
To qualify, your project needs 5,000+ GitHub stars or 1M+ monthly npm downloads, with recent commits or PR activity. Maintainers, core contributors, and anyone keeping a critical package alive are all eligible. Apply at claude.com/contact-sales/claude-for-oss. Note: this does not include API access.
|
|
|
At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 200,000 developers stay informed and ahead. We're passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.
Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We're based in the United States. By using our site, you agree to be governed by U.S. laws.
|
|
|
Looking to promote your company, product, service, or event to 250,000+ AI developers?
|
|
|
|
|