Skip to main content
Back

Sakana Fugu auto-routes AI models via one API endpoint

Sakana ships export-control bypass model, Claude joins Slack as coder, GLM-5.2 beats Opus at half cost, Bengio warns on alignment

June 24, 2026 6 min read

Hey,

The real story this week isn't any single model. It's where AI is living. Claude moves into your Slack channel as a persistent teammate. Sakana Fugu turns model selection into an invisible plumbing layer. And a new paper proves smaller, smarter training beats brute force.

The pattern: AI is quietly becoming infrastructure, not a tool you open.

And Fugu's export-control angle? That's geopolitics meeting architecture. Wild times.
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 25 sec

Top News

▸ Sakana AI ships a multi-agent orchestration model designed to bypass export controls

Tiger Data

▸ Stop Splitting Your AI Stack Across Databases

Top News

▸ Anthropic ships Claude as a Slack teammate that writes and merges your code

Lambda

▸ Optimize your AI training runs

Top Repo

▸ GLM-5.2 beats Claude Opus at real bug fixes for half the cost

Signals

1. Baidu releases OCR tool that reads entire documents in one pass
2. Presented by Viktor: Meet Viktor, the AI employee in Slack and Teams. He runs ops, finance, and outreach across 3,200+ tools
3. New theory proves self-predicting latents need exponentially less training data
4. Yoshua Bengio warns visible reward dashboards can corrupt AI safety alignment
5. Open-source 3B text-to-image model beats FLUX.1 Dev using only public data
6. Builder.io ships free open-source screen recorder built for AI agents to read
Top News
Sakana AI ships a multi-agent orchestration model designed to bypass export controls
45,885 Likes
Sakana AI ships a multi-agent orchestration model

Most AI tools make you pick one model and hope for the best. Sakana Fugu flips that. It is itself a small AI model that has learned to call other AI models, pick the right ones for each step, and stitch the results together automatically.

You hit one API endpoint. Fugu handles everything behind the scenes.

It comes in two versions:

  • Fugu: fast, low-latency, great for coding assistants like Codex and everyday chatbot work. You can also block specific models from its pool for data compliance.
  • Fugu Ultra: the heavy-hitter. Pulls in a deeper pool of expert models for hard tasks like cybersecurity analysis, research, and patent investigations. Matches top frontier models on benchmarks.

The API is OpenAI-compatible, so dropping it into existing projects is straightforward. It hits top scores on SWE-Pro, GPQA-D, and ALE-Bench.

The bigger idea: if one model gets hit by export controls and disappears, Fugu just swaps in another. No code changes on your end.

Read more
forward →
Presented by Tiger Data
Stop Splitting Your AI Stack Across Databases

As AI workloads grow, most teams end up with a database for each one.

TimescaleDB extends Postgres so one handles them all: time-series, embeddings, analytics.

No pipeline, no drift, no second database.

Get $1000 credit
partner with us →
Top News
Anthropic ships Claude as a Slack teammate that writes and merges your code
20,243 Likes
Anthropic ships Claude as a Slack teammate that wr

Anthropic just shipped Claude Tag, and it's a pretty different take on AI tooling. Instead of opening a chat window, you just type @Claude in Slack and hand it a task.

Here's what makes it click. Within a channel, there's one Claude that interacts with everyone, so anyone can pick up the conversation from where the last person left off. No more re-explaining context to a fresh chat window.

What you can actually do with it:

  • Tag @Claude with a request, and it breaks the task into stages, executes them, and posts results back in a Slack thread.
  • With ambient mode on, Claude proactively flags relevant info from across channels and follows up on threads that went quiet.
  • It builds context as it follows the channel, so you stop explaining the same background on every request.

The proof point: 65% of Anthropic's product team's code is now created by their internal version. It runs on Opus 4.8 and is available now in beta for Enterprise and Team plans.

Read more
forward →
Presented by Lambda
Optimize your AI training runs

Most AI training runs use less than half their available compute. Lambda's team found the root causes and built a playbook that cut costs >25% without changing the model.

Learn how to:

  • Find memory inefficiencies driving up costs
  • Optimize your training configuration for full hardware utilization
  • Fix GPU communication bottlenecks
Get the guide
partner with us →
Top Repo
GLM-5.2 beats Claude Opus at real bug fixes for half the cost
6,886 Likes
GLM-5.2 beats Claude Opus at real bug fixes for ha

Someone ran a real-world test: fix an actual bug from the Cline repo. Two models, same tools, same setup. Who wins?

Both GLM-5.2 and Claude Opus fixed the bug. But the details tell a very different story.

  • Cost: GLM cost $0.41 vs Opus at $0.81. Half the price.
  • Speed: Opus finished in 1.6 min with 12 tool calls. GLM took 4.7 min and 28 tool calls.
  • Quality: GLM cleaned up dead code and verified the build actually compiled. Opus left type errors that broke the production build.

Here's the interesting part. GLM used twice the tokens but still cost less. It seems trained to keep checking its own work before calling it done. Opus moved fast but shipped broken code.

For you, this means cheaper runs that actually hold up in production. Try it free today:

npm i -g cline

Read more
forward →
Signals
1
Baidu releases OCR tool that reads entire documents in one pass
3,416 Likes
2
Meet Viktor, the AI employee in Slack and Teams. He runs ops, finance, and outreach across 3,200+ tools
Presented by Viktor
3
New theory proves self-predicting latents need exponentially less training data
446 Likes
4
Yoshua Bengio warns visible reward dashboards can corrupt AI safety alignment
430 Likes
5
Open-source 3B text-to-image model beats FLUX.1 Dev using only public data
237 Likes
6
Builder.io ships free open-source screen recorder built for AI agents to read
2,886 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