Skip to main content
Back

Cursor Origin Git Platform Launch, AI Agents at 22 Commits/sec

Z.ai ships GLM-5.2 with 1M context, Cursor launches Origin git platform, Hermes agent buys via Stripe, Gemma 4 runs on 12GB VRAM

June 17, 2026 6 min read

Hey,

The coding stack is getting rebuilt from scratch, and fast.

Cursor isn't just an editor anymore. Origin is a full git hosting platform, because 22 commits per second broke GitHub's assumptions. Meanwhile, a 3B model matches Claude Opus on math. Open weights keep punching up.

We're watching the entire dev infrastructure get rewired for agents. Not adapted. Rewired.
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 41 sec

Top News

▸ Z.ai ships GLM-5.2 open-weight model with 1M context and stronger coding

Retool

▸ AI made building easy. Retool makes it safe.

Top News

▸ Cursor announces Origin, a git hosting platform built for AI agents

Gladia

▸ Real audio is noisy. Solaria-3 handles it best.

Top News

▸ Nous Research's Hermes agent can now buy things and pay APIs using Stripe

Signals

1. Google's Gemma 4 12B coding model runs locally on 12GB VRAM
2. Presented by Slack: Your team doesn't need more meetings. See 10 ways Slack can help you get more done with fewer meetings
3. Open-source 3B model matches Claude Opus 4.5 on math and coding benchmarks
4. New 3D surface model turns any number of photos into a clean mesh
5. Microsoft's FastContext model cuts coding agent token use by offloading repo search
6. Developer benchmarks Claude Code against two open-source coding tools head-to-head
Top News
Z.ai ships GLM-5.2 open-weight model with 1M context and stronger coding
7,577 Likes
Z.ai ships GLM-5.2 open-weight model with 1M conte

Z.ai just dropped GLM-5.2, an open-weights model built for serious coding work. Here is what makes it interesting.

The biggest upgrade is the context window. 1 million tokens means you can feed it an entire codebase in one shot and ask it to refactor, debug, or extend it without losing track. That jumped 5x from GLM-5.1's 200K limit.

You also get two reasoning modes to control speed vs. quality tradeoffs:

  • Max mode: pushes hardest on complex problems, uses more compute
  • High mode: faster and cheaper, still strong for most tasks
  • MIT license: download the weights, self-host it, fine-tune it, ship it commercially, no strings attached
  • Benchmark scores: 81.0 on Terminal-Bench 2.1, up from 62.0 on GLM-5.1

On pricing, it runs at $1.40 per million input tokens, roughly one-sixth of GPT-5.5. You can grab the weights on Hugging Face or hit the Z.ai API directly.

Read more
forward →
Presented by Retool
AI made building easy. Retool makes it safe.

Retool launched a new app builder to close the gap between AI and production.

Build in Retool or your preferred coding agent, then deploy to a governed foundation with auth, permissions, and audit logs built in.

Limited time offer: Double AI credits on all paid plans, plus free app imports through July 1.

Start building now
partner with us →
Top News
Cursor announces Origin, a git hosting platform built for AI agents
6,114 Likes
Cursor announces Origin, a git hosting platform bu

Cursor just announced Origin, their own git hosting platform, and it's a direct shot at GitHub.

Here's the quick backstory: Cursor (the AI code editor) acquired Graphite, a code review tool, and now they're stacking a full hosting layer on top. The goal is to own the entire coding pipeline, write in Cursor, review with Graphite, host on Origin.

So why does this matter? AI agents are now opening pull requests (proposed code changes) faster than humans can review them. GitHub was built for humans, not for dozens of AI agents committing code in parallel. A live demo showed 22.6 commits per second inside a single repo, a load GitHub was never designed for.

Origin is built specifically for that world. Key features include:

  • Scales to handle massive agent-driven commit volumes
  • Built-in merge conflict resolution handled automatically by agents
  • Extensible via API and MCP (a standard way for AI tools to talk to each other)
  • Native integration with Cursor's existing agent workflows

You can join the waitlist now. Full release is coming this fall.

Read more
forward →
Presented by Gladia
Real audio is noisy. Solaria-3 handles it best.

Most ASR models are tuned for clean speech. Datasets prove Solaria-3 was built for actual production:

  • #1 on conversational telephone audio (Switchboard)
  • #1 on business speech (Earnings22)
  • Optimized for EN, FR, DE, ES, IT. Available via the Gladia API

Try it on your own audio for free.

Try solaria-3
partner with us →
Top News
Nous Research's Hermes agent can now buy things and pay APIs using Stripe
4,729 Likes
Nous Research's Hermes agent can now buy things an

Your AI agent just got a wallet. Hermes Agent now has three official Stripe skills, and together they let your agent handle real money, on its own, with guardrails you control.

Here is what you can install:

  • stripe-link-cli -- lets your agent buy things on the open web, like a human clicking checkout
  • mpp-agent -- lets your agent pay for APIs that charge per call, so it can keep working without stopping to ask you for a credit card
  • stripe-projects -- lets your agent spin up and bill its own SaaS products, no human required

To get any of these, run: hermes skills install official/payments/[skill-name]

The safety story matters here. Your real credentials never appear in the agent's conversation history. You can require human approval before any payment fires. One-time credentials are deleted after use. Stripe Projects issues scoped credentials tied to a specific project, with pre-defined permissions and spending guardrails.

What this opens up: agents that fund themselves, pay for their own tools, and ship products, all while you stay in control of the limits.

Read more
forward →
Signals
1
Google's Gemma 4 12B coding model runs locally on 12GB VRAM
2,618 Likes
2
Your team doesn't need more meetings. See 10 ways Slack can help you get more done with fewer meetings
Presented by Slack
3
Open-source 3B model matches Claude Opus 4.5 on math and coding benchmarks
1,655 Likes
4
New 3D surface model turns any number of photos into a clean mesh
532 Likes
5
Microsoft's FastContext model cuts coding agent token use by offloading repo search
192 Downloads
6
Developer benchmarks Claude Code against two open-source coding tools head-to-head
2,569 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 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