Skip to main content
Back

Cognition $1B Raise , Google Universal Embeddings , C++ Body Track

Cognition raises $1B at $26B, Google DeepMind ships universal embedding model, Qwen3 8B runs agent on old GPU, KAIST optimizer trains 1.5x faster

May 28, 2026 6 min read

Hey,

AI is eating its own cooking. Devin now writes 89% of Cognition's code, and the market just rewarded that with a $1B raise at a $26B valuation. Revenue went from $37M to $492M in a year.

Meanwhile, Google quietly dropped a single embedding model for text, audio, video and images. The infrastructure layer is consolidating fast.

The pattern is clear: agents are becoming the product, not the feature. We're watching software development restructure itself in real time.
Together 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 49 sec

Top News

▸ Cognition raises $1B at $26B valuation as Devin writes 89% of its own code

Buildkite

▸ How frontier AI labs ship: it starts at CI

Top Model

▸ Open-source C++ body tracking engine maps 70 joints in real time, no Python needed

Slack

▸ The Context Opportunity: Agentic AI at Scale

Top Paper

▸ Google DeepMind releases one embedding model for text, audio, video, and images

Signals

1. Qwen3 8B runs a full coding agent at 20 tok/s on a 10-year-old GPU
2. Presented by Braintrust: Braintrust explains how to evaluate multi-turn conversations for more complex AI products
3. Crawl4AI ships an open-source web scraper built specifically for LLMs
4. KAIST's new optimizer trains LLMs 1.5x faster with no learning rate schedule
5. OpenBMB launches $100K contest to push 9B model to 1M-token inference
6. Micrograd builds a working neural net in just 150 lines of Python
Top News
Cognition raises $1B at $26B valuation as Devin writes 89% of its own code
2,515 Likes
Cognition raises $1B at $26B valuation as Devin wr

Cognition just raised $1B at a $26B valuation for Devin, its AI coding agent. That valuation more than doubled from $10.2B just eight months ago. The revenue story is even wilder: run-rate grew from $37M to $492M in twelve months.

So what actually is Devin? It picks up a task, writes the code, tests it, and opens a pull request, all without you touching it. Think of it as a teammate that never sleeps and handles the boring tickets.

Here's what's new and what it unlocks for you:

  • Devin now writes 89% of Cognition's own code, up from 13% in December. They eat their own cooking.
  • Their new model SWE-1.6 runs at up to 950 tokens per second in Windsurf, making it fast enough to feel like a real pair-programming partner.
  • Cognition stays model-agnostic, routing tasks to the best AI model available, so you get better results without managing that complexity yourself.
  • Mercedes-Benz cut an eight-month legacy modernization project down to eight days using Devin.

The takeaway: autonomous coding agents just crossed from experiment to production reality.

Read more
forward →
Presented by Buildkite
How frontier AI labs ship: it starts at CI

Your CI should get out of the way.

Every layer of the AI stack runs CI on Buildkite: frontier labs (OpenAI, Anthropic, Mistral, Cohere); inference (vLLM, Modular, Anyscale, Groq); ML platform infra (Hugging Face, Tecton, Anomalo); applied (Harvey, Persona). Hybrid agents on your GPUs. Your secrets, your scale, while we orchestrate the rest.

Buildkite won't buckle. We're running 1.3B job mins/wk with a sea of green on our statuspage and room for more.

Our trial unlocks everything. 30 days, all access.

Make ship happen
partner with us →
Top Model
Open-source C++ body tracking engine maps 70 joints in real time, no Python needed
1,482 Likes
Open-source C++ body tracking engine maps 70 joint

Most 3D body tracking tools need Python and heavy ML frameworks just to get started. SAM3DBody-cpp skips all of that.

It's a pure C++ engine that reads a camera feed and instantly outputs 70 joint positions across the full body and both hands, plus a full 3D mesh of the person. No Python. No PyTorch. Just fast, native code.

Under the hood, it runs Meta's SAM 3D Body model, which estimates human pose of the body, feet, and hands using a mesh representation that decouples skeletal structure and surface shape. The C++ wrapper makes that power actually usable in real products.

What you can build with it:

  • Robot perception that reacts to human movement in real time
  • Motion capture pipelines without expensive hardware
  • Games or AR apps that track your whole body live

The API is a lightweight C interface, so you can call it from any language. Models are on Hugging Face, code is on GitHub.

Read more
forward →
Presented by Slack
The Context Opportunity: Agentic AI at Scale

AI agents work best when they have context, but most enterprises still force agents to ask for details before acting.

Slack serves as the enterprise conversational layer that gives agents the context they need to act intelligently, at scale, right in the flow of work.

Read the report
partner with us →
Top Paper
Google DeepMind releases one embedding model for text, audio, video, and images
574 Likes
Google DeepMind releases one embedding model for t

Imagine you have a search engine that understands text, images, audio, and video all at once. That's what Gemini Embedding 2 does.

An embedding model converts any input into a list of numbers that captures its meaning. Similar things get similar numbers. This is how search, recommendations, and AI memory work under the hood.

Most models only handle one type of input. GE 2 handles all four in a single unified model, so you can search across types. Ask with an image, get back a video. Mix and match freely.

  • Tops benchmarks on image retrieval, video search, multilingual text, and code
  • Works out-of-the-box on niche domains like astronomy or culinary arts it was never trained on
  • Beats Google's previous text-only models on text tasks
  • Available now on Gemini API and Vertex AI

This opens up cross-modal RAG (where your AI assistant retrieves relevant info to answer questions) using any media type, smarter recommendation systems, and richer semantic search. One model, every format.

Read more
forward →
Signals
1
Qwen3 8B runs a full coding agent at 20 tok/s on a 10-year-old GPU
308 Likes
2
Braintrust explains how to evaluate multi-turn conversations for more complex AI products
Presented by Braintrust
3
Crawl4AI ships an open-source web scraper built specifically for LLMs
66,571 Stars
4
KAIST's new optimizer trains LLMs 1.5x faster with no learning rate schedule
209 Likes
5
OpenBMB launches $100K contest to push 9B model to 1M-token inference
9,146 Stars
6
Micrograd builds a working neural net in just 150 lines of Python
16,040 Stars
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?
Work with us