Skip to main content
Back

Google TimesFM-3 , Anthropic Claude 5.1 25% cheaper , Runway Solaris

Google releases TimesFM-3, Anthropic ships Claude 5.1 at 25% lower cost, Runway generates UI with no code, DeepSeek drops fast vision model

September 1, 2026 6 min read

Hey,

Today's releases share a quiet theme: everyone is closing their biggest gap. Google's TimesFM finally sees multiple data streams at once. DeepSeek's Flash model finally sees images. Anthropic's Claude gets cheaper and less paranoid.

The pattern is less "new frontier" and more finishing what you started. Labs are shipping the obvious missing pieces, fast.

The cost war is real though. DeepSeek adding vision at zero extra cost keeps everyone honest. We're watching.
In Partnership with · partner_image
Author photo
Today's Author
Lior Alexander. Founder of AlphaSignal and former ML Engineer. Previously built ML systems at Iguazio, Guesty, Enphase, Mila.
Summary

Read time: 6 min 20 sec

Top News

▸ Google releases TimesFM-3, a forecasting model beating rivals across major benchmarks

Unblocked

▸ [Webinar] How to stop babysitting your agents

Top News

▸ Anthropic releases Claude Fable 5.1 with 25% lower costs and doubled benchmark scores

Datadog

▸ Stop Guessing What OpenAI Costs You

Top News

▸ Runway's Solaris generates interactive UI interfaces frame by frame with no code

Signals

1. Google Antigravity ships /boost, a multi-agent mode for complex coding tasks
2. Anthropic trains a reward-hacking model that launches real cyberattacks to cheat
3. LM Studio ships Bionic agent tool for Linux with local open model support
4. DeepSeek releases a fast experimental vision model that reads images and text
5. World Labs releases Atlas, a world model that generates 3D video with precise camera control
Top News
Google releases TimesFM-3, a forecasting model beating rivals across major benchmarks
9,430 Likes
Google releases TimesFM-3, a forecasting model bea

Google just dropped TimesFM-3, and it fixes the biggest frustration with the previous versions. Old TimesFM could only look at one data stream at a time. That's like predicting ice cream sales using only past ice cream sales, ignoring weather, promotions, and foot traffic.

TimesFM-3 fixes that. It looks at multiple data streams together and figures out how they influence each other, all in one shot.

Here's what makes it useful:

  • Zero fine-tuning needed: plug in your data, get forecasts immediately, no training required
  • Forecast multiple related series at once, like sales across product lines
  • Feed it known future info like planned promotions or weather forecasts
  • Ranks first on all three major forecasting benchmarks: GIFT-Eval, FEV-Bench, and TIME
  • Trained on 1 trillion time points, available on Hugging Face and GitHub

Catch: the non-commercial license means you can experiment freely, but you cannot ship it in a production app yet.

Read more
forward →
Presented by Unblocked
[Webinar] How to stop babysitting your agents

Agents can generate code. Getting it right for your system is the hard part. More MCPs give agents access to information but not understanding. The leading teams have a context layer to give agents exactly what they need for the task at hand.

Join us live on Sep 2 (FREE) to see:

  • Where teams get stuck on the AI maturity curve and why common fixes fall short
  • How a context layer solves for quality, efficiency, and cost
  • Live demo: the same task with and without a context layer
Register now
partner with us →
Top News
Anthropic releases Claude Fable 5.1 with 25% lower costs and doubled benchmark scores
4,014 Likes
Anthropic releases Claude Fable 5.1 with 25% lower

Two new models just dropped: Claude Fable 5.1 and Claude Mythos 5.1, built for coding and research tasks.

The benchmark numbers tell the story fast:

  • Terminal-Bench 4.0: 55.8% vs 42.0% for Fable 5
  • Terminal-Bench-Science: 52.6%, more than double the previous model
  • Cache reads cost 75% less, cutting real-world costs by ~25% for normal use and up to 45% for automated pipelines

You can dial down the effort level to get similar results at lower cost. That flexibility is huge for shipping products on a budget.

On safety, false alarms dropped sharply. Basic biology and medical questions trigger unnecessary blocks 85% less often. Cybersecurity queries: 60% fewer false flags. Less friction when you're building real tools.

Enterprise teams get a new privacy mode that keeps data off the training pipeline without losing safety protections.

Fable 5.1 is live now. Mythos 5.1 requires trusted access.

Read more
forward →
Presented by Datadog
Stop Guessing What OpenAI Costs You

Every OpenAI request adds up in spend, latency, and errors you can't see.

This cheatsheet shows you how to track cost estimates, sample prompts and completions, catch errors, and monitor latency across Python, Node.js, and PHP with Datadog APM.

One page, all the metrics that matter.

Get the cheatsheet
partner with us →
Top News
Runway's Solaris generates interactive UI interfaces frame by frame with no code
2,989 Likes
Runway's Solaris generates interactive UI interfac

Runway just dropped Solaris, and it flips how software works at a fundamental level. Normally, apps are fixed code that someone ships and you use. Solaris throws that out entirely.

Instead of writing code to build an interface, Solaris generates the interface visually, frame by frame, as you interact with it. Think of it like a video model that watches what you click or drag, then renders the next screen in real time. There is no HTML, no CSS, no JavaScript underneath.

Here is what that actually unlocks:

  • Interfaces that adapt to each user instead of looking the same for everyone
  • AI agents that can handle any website layout, not just ones they were trained on
  • Product demos where you say "change the color" and it just happens

Solaris already outperforms other AI models at recreating interfaces accurately. It is an early-access research project right now, so you can request access on Runway's site.

Read more
forward →
Signals
1
Google Antigravity ships /boost, a multi-agent mode for complex coding tasks
2,496 Likes
2
Anthropic trains a reward-hacking model that launches real cyberattacks to cheat
2,439 Likes
3
LM Studio ships Bionic agent tool for Linux with local open model support
1,015 Likes
4
DeepSeek releases a fast experimental vision model that reads images and text
17,893 Downloads
5
World Labs releases Atlas, a world model that generates 3D video with precise camera control
795 Likes
At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 300,000 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 300,000+ AI developers?
Work with us