A Glimpse at GPT-5.6 Sol’s Mathematical Breakthroughs

Advertise here | 6-min Read

WHAT’S HAPPENING AI TODAY

1. MiniMax dropped H3 open weights — but excluded US, EU, UK, and South Korea from the license: MiniMax Drops H3 Open Weights With License That Excludes US, EU, UK and South Korea. H3 is MiniMax's frontier video model with native 2K resolution and stereo audio — and its open weights are now available, but not for anyone in the four largest AI markets. This is the first major open-weight AI release with explicit geographic exclusions targeting Western markets. The license structure means teams in the US, EU, UK, and South Korea cannot legally use the weights for any purpose. For teams in other markets: H3 at open weights and $0.08 per minute for the API is the most capable video model available at that price point. For Western teams: watch for a commercially licensed variant, or evaluate the API-only access that remains available globally.

2. CrowdStrike: AI-enabled adversary attacks jumped 89% — and DPRK poisoned 131 npm packages: CrowdStrike Threat Hunting Report: AI-Enabled Adversary Attacks Jumped 89%, DPRK Poisoned 131 npm Packages. The 89% jump in AI-enabled attacks is not a projection. It is CrowdStrike's measured year-on-year figure from its threat hunting operations. North Korea's 131 npm packages are a supply chain attack that targeted developer toolchains directly — the same vector that the jscrambler malware used in June against Claude Desktop, Cursor, and VS Code configs. For any team running Node.js in production: audit your npm dependency tree against the CrowdStrike IOCs published today.

3. OpenAI gave 100,000 academic researchers free access to GPT-5.6 Sol Pro — starting with 10,000 this summer: OpenAI is launching ChatGPT for Academic Researchers, a free program giving 100,000 scientists, mathematicians, and engineers access to frontier models including GPT-5.6 Sol Pro. Starting with 10,000 researchers this summer at institutions like the Institute for Advanced Study and École normale supérieure, the program expands through 2027. The same week OpenAI's Astra solved ten open math problems, it is making frontier model access free to the researchers most likely to find the next ten. The scientific AI strategy is now explicit: solve problems, then give scientists the tools to solve more.

🏛️ Anthropic, OpenAI, Google, Meta head to the White House — today

Executives from Anthropic, OpenAI, Google, and Meta arrived at the white house today for the first time to see how the US government intends to inspect their most powerful models before release. The meeting was convened by the Office of the National Cyber Director and includes National Cyber Director Sean Cairncross, Treasury Secretary Scott Bessent, and Commerce Secretary Howard Lutnick. The framework is now complete — labs are seeing it for the first time today.

The details:

• The framework springs from Trump's June 2 executive order on AI cybersecurity. It sets out how officials will assess whether frontier models can find and exploit software vulnerabilities — the capability that triggered the Fable 5 ban. Participation is voluntary. The EO explicitly bars the process from becoming a licensing, permitting, or preclearance regime.

• The EU AI Act went into effect on August 2 — one day before this meeting. The European Commission can now demand to review AI models before release and impose fines of up to 7% of annual gross revenue for non-compliance. US and EU governance frameworks are now both active simultaneously, with different requirements and enforcement mechanisms.

• The administration has not yet designated a single person or organisation within the White House to lead ongoing AI company outreach — multiple sources within the AI companies told CNN. The framework exists; the institutional home for it does not yet.

• Today's meeting is the first time the labs will know the precise criteria for what qualifies as a "covered frontier model" subject to review. Until today, that threshold was classified.

Why it matters: The Fable 5 ban happened because there was no framework — just a bilateral crisis response. Today's meeting is the beginning of the alternative. The voluntary framework does not prevent another Fable 5. But it creates the infrastructure for labs to engage with the government before a crisis rather than during one. For any team building on frontier models: the governance layer above the model layer is now being formalised.

A Glimpse at GPT-5.6 Sol’s Mathematical Breakthroughs

Decide whether to upgrade to Claude Opus 5 — before your next billing cycle

Prompt: You are a senior security engineer. CrowdStrike's Threat Hunting Report confirmed today that AI-enabled adversary attacks jumped 89% year-on-year. In the same report: North Korea poisoned 131 npm packages in a supply chain campaign targeting developer toolchains — the same vector used in June when jscrambler malware targeted Claude Desktop, Cursor, Windsurf, VS Code, and Zed configuration files.

The pattern is consistent: attackers are targeting the tools developers use to build with AI, not the AI models themselves. If your pipeline installs npm packages automatically — in CI/CD, in Docker builds, in agent environments — you have exposure.

Help me run a supply chain security audit across four areas:

1. The dependency inventory — list the categories of npm packages in our production and CI/CD environments that carry the highest supply chain risk. For each category: what is the attack pattern (typosquatting, dependency confusion, maintainer compromise), and what is the one technical control that most reduces exposure for that specific attack pattern?

2. The DPRK IOC check — CrowdStrike published indicators of compromise for the 131 poisoned packages today. Write the three shell commands I should run right now to check whether any of those package names appear in our package.json, package-lock.json, or installed node_modules. Make them specific and runnable, not pseudocode.

3. The AI toolchain exposure — our team uses [list your AI coding tools — Cursor, Claude Code, Copilot, Windsurf, etc.]. For each tool: does it install or execute npm packages automatically, and if so, in what context? The jscrambler attack targeted config files. The DPRK attack targeted packages that get auto-installed. Where do these two vectors intersect in our setup?

4. The lockfile policy — the most effective single control against supply chain attacks is committing package-lock.json and using npm ci instead of npm install. Is this enforced in our CI/CD pipeline? If not, write the three-line change to our pipeline config that enforces it starting today.

End with a go/no-go for deploying any new npm packages this week — given the active DPRK campaign — and the specific condition that would change that answer.

AI news highlights

• MiniMax H3 open weights — excludes US, EU, UK, South Korea from the license — the first major open-weight release with explicit Western market exclusions. API access remains global at $0.08/min.
• AI-enabled adversary attacks up 89% — DPRK poisoned 131 npm packages — CrowdStrike Threat Hunting Report. Audit your npm dependency tree against today's IOCs if you run Node.js in production.
• OpenAI gave 100,000 researchers free GPT-5.6 Sol Pro access — starting this summer — IAS, ENS, and others. The week after Astra solved open math problems, OpenAI gives scientists the tools to find the next ones.
• xAI launched Grok Voice Think Fast 2.0 — becomes grok-voice-latest on August 5 — improved speech-to-speech AI. Priced at $0.08 per minute, same as MiniMax H3 video. Voice AI pricing is converging.
• Google Gemini free video trial — 10 videos at no cost, ends tonight at 11:59pm PT — available to users without an existing Google AI subscription. Powered by Gemini Omni. If you want to test AI video without committing, today is the window.
• UK photonic chip startup OLIX raised $312M at $3.3B — backed by Arm and Reed Hastings — light-based AI compute, not silicon. The bet: photonic chips cut AI inference energy costs by orders of magnitude. If it works, the economics of frontier inference change permanently.
• NVIDIA $250B OpenAI backstop is in advanced talks — Piketon, Ohio, former uranium site — 10 gigawatts. The compute market is concentrating faster than the model market. Watch this close.
• Claude Sonnet 5 introductory pricing: 28 days left at $2/$10 — standard $3/$15 from September 1. Lock in your routing decisions before the switch.

Trending AI tools

Research & Content

• Perplexity AI — AI-powered search with fact-checked answers and direct citations
• Gemini Notebook — Upload docs, generate audio overviews, study guides, and direct answers
• Claude — Thoughtful, context-heavy writing and deep document analysis

Software Development

• Cursor — AI-powered code editor built on VS Code — build software with agents and natural language
• v0 by Vercel — Turn text prompts into functional React and Tailwind code instantly

Video & Audio

• ElevenLabs — Industry standard for text-to-speech, voice cloning, and audio generation
• Synthesia — Scripts to professional corporate videos using photorealistic AI avatars
• Runway — Realistic video clips from text and image prompts

Productivity & Automation

• Notion AI — Writing, summarising, and workspace organisation in one second-brain interface
• Gamma — Notes or prompts to polished presentations, documents, and webpages instantly

That’s a Wrap

SPONSOR US

Get your business in front of over 90k+ AI professionals

8020AI is the world’s #1 AI Newsletter, Read by 90k+ professionals from leading companies such as Google, OpenAI, Meta, and Microsoft.

We've assisted in promoting Over 500 AI-Related Products. Will yours be the next?

What We Can Offer:

• Launch an Advertising Campaign

• Introduce New Product or Features

• Other Business Cooperation

Get in touch today.

Or Email our founder Alamin at [email protected]

FEEDBACK

How was your experience with 8020AI today?

Login or Subscribe to participate in polls.

If you have specific feedback or anything interesting you’d like to share, please let us know by replying to this email.