Skip to main content
Back

Anthropic Mythos 5 back for 100 orgs while Fable 5 stays locked

OpenAI previews GPT-5.6 family, Nous Research beats Claude on benchmarks, LLMs reuse 11 words in 88% of stories, Codebase-memory indexes Linux in 3min

June 29, 2026 6 min read

Hey,

The US government is now a de facto gatekeeper for frontier AI. Both OpenAI and Anthropic launched their most powerful models this week, and both got the same answer: not so fast.

GPT-5.6 Sol, Terra, Luna. Claude Mythos 5 back for ~100 orgs. The pattern is clear: the more capable the model, the more Washington wants a seat at the table.

We're watching a new access tier being invented in real time. Not free, not paywalled. Government-walled.
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 44 sec

Top News

▸ OpenAI previews GPT-5.6 family with three models spanning speed, cost, and power

CopilotKit

▸ Any agent on any platform.

Top News

▸ Anthropic restores its top cybersecurity model to US critical infrastructure operators

TokenGo

▸ TokenGo Inference, building the intelligence grid

Top News

▸ Nous Research ships gated model blends that beat Claude and GPT-5.5 on benchmarks

Signals

1. OpenAI ships Codex app update with faster scrolling and smarter navigation
2. Presented by AssemblyAI: The first realtime speech-to-text that takes your agent's question as input. 6.99% WER, 18 languages, ~300ms. Still $0.45/hr
3. NumPy remains the go-to open-source math library powering Python science
4. New free book covers the full stack of building agentic AI systems
5. Study finds LLMs recycle the same 11 words in 88% of generated stories
6. Codebase-memory indexes the Linux kernel in 3 minutes, cutting agent tool calls by 2x
Top News
OpenAI previews GPT-5.6 family with three models spanning speed, cost, and power
60,511 Likes
OpenAI previews GPT-5.6 family with three models s

OpenAI just dropped a new model family: GPT-5.6 Sol, Terra, and Luna. Three models, one for every budget and use case.

Here is what each one is for:

  • Sol is the flagship. Best for hard problems: complex coding, security research, long multi-step tasks. Priced at $5 input / $30 output per 1M tokens.
  • Terra matches GPT-5.5 quality at 2x lower cost ($2.50/$15). Swap it in for your current GPT-5.5 workloads and save money immediately.
  • Luna is the cheapest at $1/$6. Built for high-volume, fast tasks like summarization and routine automation.

Sol topped Terminal-Bench 2.1, a test for complex command-line workflows requiring planning and tool use. It also excels at vulnerability research and cybersecurity tasks.

One catch: the U.S. government asked OpenAI to limit the rollout to roughly 20 trusted partner organizations for now. Broader API and ChatGPT access is coming in the next few weeks.

Read more
forward →
Presented by CopilotKit
Any agent on any platform.

Your users already live in Slack, Teams, Telegram, WhatsApp and Discord.

Now your agent can live there too.

CopilotKit makes it real: one backend on the open AG-UI protocol, five native bots.

The same agent that streams its work, syncs state, and pauses for human approval now lives inside the apps your team uses all day.

Pick any agent framework, swap it whenever.

The bots and the protocol stay the same, so there's no per-platform rewrite.

Deploy to chat
partner with us →
Top News
Anthropic restores its top cybersecurity model to US critical infrastructure operators
30,123 Likes
Anthropic restores its top cybersecurity model to

Anthropic's two most powerful AI models got pulled offline two weeks ago under a US government directive. Here's where things stand now.

Claude Mythos 5 is Anthropic's strongest cybersecurity model. It can autonomously find software vulnerabilities at a scale that genuinely spooked the government. Think: thousands of high-severity flaws across every major OS and browser, including a 27-year-old OpenBSD bug and chained Linux kernel exploits.

That power is exactly why it got locked down. The concern: bad actors could use it to attack systems just as fast as defenders use it to protect them.

Now, Mythos 5 is partially back. Access is authorized for roughly 100 trusted US organizations, primarily cybersecurity companies and critical infrastructure providers.

  • Project Glasswing members like Apple, Google, Cisco, Nvidia, and Microsoft are cleared to resume access.
  • API users and international customers are still limited to older Claude models.
  • Fable 5, the general-use version, remains offline for everyone else.

Anthropic is actively negotiating to bring Fable 5 back for general use. No access yet for most of us, but the door is opening.

Read more
forward →
Presented by TokenGo
TokenGo Inference, building the intelligence grid

TokenGo operates and optimises inference at below market rates with no quality sacrifice.

10% off market rates across the board to start. One base URL change, same SDK. Further volume-based tiered discounting available.

Contact our team to unlock $100 in free credits or start testing in seconds with TokenGo LLM API.

Get started
partner with us →
Top News
Nous Research ships gated model blends that beat Claude and GPT-5.5 on benchmarks
7,118 Likes
Nous Research ships gated model blends that beat C

The best AI models are locked behind waitlists. Most of us can't touch them. Hermes Agent just built a workaround.

It's called Mixture of Agents (MoA): instead of one model answering your question, several models tackle it at once. Each one responds independently. Then a final "aggregator" model reads all those answers and writes one clean output. Multiple brains, one response.

Numbers from their internal benchmark:

  • 8% better than Claude Opus 4.8 running solo
  • 11% better than GPT-5.5 running solo
  • A two-model combo scored 82% vs 76% for Opus alone

To use it: run hermes model and pick a MoA preset, or trigger it inline with /moa. Your preset shows up like any normal model. Memory, tool calls, and long sessions all still work. You can also swap presets per task using slash commands. No gated access required.

Read more
forward →
Signals
1
OpenAI ships Codex app update with faster scrolling and smarter navigation
4,872 Likes
2
The first realtime speech-to-text that takes your agent's question as input. 6.99% WER, 18 languages, ~300ms. Still $0.45/hr
Presented by AssemblyAI
3
NumPy remains the go-to open-source math library powering Python science
32,385 Stars
4
New free book covers the full stack of building agentic AI systems
1,739 Likes
5
Study finds LLMs recycle the same 11 words in 88% of generated stories
1,182 Likes
6
Codebase-memory indexes the Linux kernel in 3 minutes, cutting agent tool calls by 2x
4,370 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