Anthropic had a big day. A new Claude topping coding benchmarks and agents that finally run while you sleep. Two very different announcements, same underlying bet: Claude stops being a tool you prompt and becomes infrastructure you deploy.
The benchmark race is fun to watch, but the agents story is the real one. Cron schedules, credential vaults, parallel subagents. That's a backend engineer on autopilot.
Meanwhile, someone finally wrote the math textbook explaining why any of this works. Good timing.
|
|
In Partnership with
|
·
|
|
|
|
|
|
Today's Author
Lior Alexander. Founder of AlphaSignal and former ML Engineer. Previously built ML systems at Iguazio, Guesty, Enphase, Mila.
|
|
Top News
|
|
|
|
Cursor ships Claude with 72.9% on its coding benchmark, 8 points ahead
|
|
6,028 Likes
|
|
Claude Fable 5 just landed in Cursor, and the numbers are hard to ignore. It hit 72.9% on CursorBench, a test built from real, messy, multi-file coding tasks pulled from actual Cursor sessions. That's 8 points above the previous best. So what makes it different? It's built to stay focused across very long tasks without losing track of what you asked it to do. It holds intent across long sessions and drives tasks to completion with fewer check-ins. Think: refactoring a whole module, not just one function. Here's what it can do inside Cursor: - Search your codebase by meaning, find files by name, read directory structures, and grep for patterns
- Suggest edits, apply them automatically, and execute terminal commands
- A thinking variant is available for problems that reward extra deliberation
The tradeoff: it costs about twice as much as Claude Opus 4.8, at $10 per million input tokens and $50 per million output tokens. Use Fable 5 for the hard stuff, not everything.
|
|
|
|
Presented by Lambda
|
|
How to push Model FLOPS Utilization past 60%
|
|
Most large-scale training runs operate at 35–45% Model FLOPS Utilization (MFU). You're paying for twice the compute you're using. Lambda's engineers benchmarked Llama 3.1 models from 8B to 405B on NVIDIA Blackwell GPUs and traced efficiency loss to its root causes: - Memory overhead
- Parallelism strategy
- Serialized communication
The result: a reproducible framework that pushed MFU past 60% (25%+ improvement vs. industry benchmark) with no changes to model architecture.
|
|
|
|
Top News
|
|
|
|
Anthropic ships scheduled Claude agents with secret API key vaults and parallel subagents
|
|
3,760 Likes
|
|
Claude agents used to die when you closed the tab. That changes now with three new features. Here is what shipped: - Scheduled deployments (public beta): Give an agent a cron schedule, and each time it fires, the agent starts a new session and completes its task with no scheduler for you to build or host. Perfect for nightly data syncs, weekly compliance scans, or daily digests.
- Environment variables in vaults (public beta): The agent never sees your real API key. The sandbox holds only a placeholder, and the real key is attached at the network boundary, only on requests to domains you allow.
- Dynamic workflows in Claude Code (now GA): Claude writes its own orchestration program on the fly, then runs it across many separate agents, each with a clean context window and one focused job.
One team deleted all the scheduling infrastructure they had built in-house and replaced it with scheduled deployments, simplifying their stack. That is the kind of boilerplate this kills.
|
|
|
|
Presented by ghost
|
|
Your Agent Needs More Than 2 Projects
|
|
Most Postgres providers give you a handful of databases and call it a free tier. Your agent burns through that before lunch. Ghost is built for how agents use databases: Fork prod in 58ms. Spin up a fresh instance per task. Delete without cleanup. 100 hours/month free. 1TB storage. No credit card.
|
|
|
|
Top News
|
|
|
|
Luma ships Ray3.2, a video model that turns text directions into cinematic shots
|
|
3,684 Likes
|
|
Luma AI just dropped Ray3.2, and the pitch is simple: you direct, the AI shoots. You're no longer just typing prompts and hoping for the best. You're actually controlling the scene. The big question: what changed? A lot. Here's what you can now do: - Up to 16 keyframes per clip, so you control exactly how motion and pacing unfold, frame by frame
- Track up to 8 faces at once, preserving expressions, posture, and gestures across the whole shot
- Export native 16-bit HDR in EXR format, which drops straight into color grading and VFX pipelines
- Reframe shots after the fact, swap backgrounds, change aspect ratios, without reshooting anything
- Generate clips up to 20 seconds at 1080p, long enough to build real scenes
The other big thing: Ray3.2 now has an API. For the first time, you can plug Luma's video generation directly into your own tools, render farms, or custom pipelines. No middleware needed.
|
|
|
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?
|
|
|
|
|