Ten AI stories moved today. The biggest one is a scorecard, and it asks every reader a blunt question. Does Monday's record spending still trust that an AI agent can do the work it is being paid to do?
A new test says the honest answer sits far short of yes. Two voices read that gap in opposite directions. Nine more stories test the same nerve: keep spending, or wait for the score to catch up.
SemiAnalysis Humbles Every Frontier Agent
A research firm built the hardest AI test yet. SemiAnalysis released GDPx today. The test grades 13 models on 206 tasks across 11 fields where real money changes hands. Those fields include chip design, cybersecurity, biology, finance and legal work1.
Claude Opus 5.5 finished first. Its score was 12.0 percent. Google's Gemini 4 Carbon took second at 10.9 percent, and OpenAI's GPT-6 Astra came third at 9.6 percent1.
Finance and legal are the two fields closest to a paycheck. Both returned a zero score for every model tested under the strict grading rule1. Dylan Patel's firm called the result a floor, still early. "We believe the first trillion dollars of AI revenue will be concentrated in sectors that can deploy frontier intelligence at high ROI," SemiAnalysis wrote.
Claude Opus 5.5 Leads SemiAnalysis's New Test at Just 12 Percent
GDPx benchmark scores, percent of 206 tasks solved across 11 valuable domains; higher is better, Nov. 2, 2026
Each bar is one model's overall GDPx score; the magenta bar is the leader, and the gray bars trail it.
Source: SemiAnalysis [1]. Chart by Ryan Elliott Dennis.
The numbers behind this chart
| Item | Value |
|---|---|
| Claude Opus 5.5 | 12% |
| Gemini 4 Carbon | 10.9% |
| GPT-6 Astra | 9.6% |
Sayash Kapoor, who studies how AI agents fail at Princeton, reads the deeper problem differently. He and five colleagues tested agents directly, and reliability barely moved even as scores climbed. "While rising accuracy scores on standard benchmarks suggest rapid progress, many agents still continue to fail in practice," Kapoor and his co-authors wrote2. AI Lately's Technology column opens that paper in full today.
OpenAI's Ask Shrinks to a Third
OpenAI wanted $30 billion for its newest funding round. Investors gave it $11 billion3.
The company cut its target to $20 billion and its valuation goal to $1.25 trillion, down from $1.4 trillion. Large investors showed only tepid interest. OpenAI now plans to court family offices and rich people for the rest3.
OpenAI's Round Lands at $11 Billion, a Third of Its First $30 Billion Ask
OpenAI's funding target and commitments secured, billions of dollars, as of Nov. 2, 2026
Each column is a dollar figure from OpenAI's own round: the first target, the lowered target, and the magenta column, what investors actually committed.
Source: TechCrunch [3]. Chart by Ryan Elliott Dennis.
The numbers behind this chart
| Item | Value |
|---|---|
| First target | $30B |
| Lowered target | $20B |
| Secured so far | $11B |
Anthropic Drops Its Five-Hour Leash
Claude Code users on Anthropic's $200 plan got a break Friday. Anthropic renamed the plan from Max 20x to Ultra. It dropped the rolling five-hour usage cap for verified accounts. A weekly cap still applies.
Current subscribers moved to the new name on their own, and the $200 price stayed exactly where it was.
Aramco Buys Into Saudi Arabia's AI Bet
A deal rare among oil companies closed today. Aramco paid roughly $2.6 billion in cash, plus AI and computing assets. The money buys a 49 percent stake in HUMAIN, Saudi Arabia's state-backed AI venture4. Saudi Arabia's Public Investment Fund keeps the other 51 percent.
Amin Nasser, Aramco's chief executive, calls the stake core strategy. "This partnership with PIF strengthens Aramco's position as a global technology leader and AI powerhouse, while laying the foundation for localising the AI industry and enabling faster, more integrated and scalable deployment of AI systems," Nasser said4. HUMAIN plans roughly three gigawatts of AI data center capacity by 2030.
DeepSeek Lets an Agent Loose on GitHub
China's DeepSeek open-sourced a new coding agent called dsh-superagent on Friday. It watches one GitHub repository and the team's chat. The agent opens pull requests two ways. An allowed person asks for the work, or a rule the team wrote already covers it5.
DeepSeek Harness@DeepSeekHarness
DeepSeek calls the release research code, a step short of a finished product5. Its GitHub repository, free under an MIT license, had drawn 11.1 thousand stars as of Monday10.
Ternus Defends Apple's Slow Walk
John Ternus, Apple's chief executive, used a WIRED interview to answer Apple's AI critics directly. Ternus argued a feature should earn its place through what it does, apart from the label stamped on it. "We don't want to ship AI for AI's sake," Ternus said. "When we're doing it right, people won't have to think about whether it was AI that made this happen or more traditional software," he told WIRED's Steven Levy6.
Suleyman Draws a Line at Consciousness
Mustafa Suleyman wants a rule set before anyone needs it. Microsoft's AI chief posted Sunday that a conscious AI system should stay unbuilt, whatever the technology allows. "If it is possible, based on everything I know, developing conscious AIs would be reckless, and we should not build them," Suleyman wrote7.
Suleyman called for an immediate pause on any AI work the moment real signs of mind appear. He wants the finding told to everyone, right away.
Altman Flies to Seoul for Memory
Sam Altman landed in South Korea on short notice. He met roughly 50 Samsung executives in Yongin on Saturday. Altman then had lunch with SK Group Chair Chey Tae-won and leaders from SK Hynix and SK Telecom8.
Altman pressed the point that AI alone lost its edge. "Simply using AI will no longer be enough to secure a sustainable competitive advantage," Altman told the Samsung room8. Both sides stayed quiet on a new supply deal or price.
Nobel Winners Keep Walking Out
Top scientists keep trading the biggest labs for something smaller. AI Lately's Talent column asks why today, with a sharper question: this time about science-focused AI startups.
Animators Win a Raise and a Rule
SAG-AFTRA members ratified new television animation contracts this week, 94.95 percent in favor9. The deal lifts voice actor pay 12.6 percent over four years and adds new rules on AI-generated performances, retroactive to July 1.
Duncan Crabtree-Ireland, the union's national executive director and chief executive officer, ties the raise to the same cause. "SAG-AFTRA members who work in TV Animation are entitled to meaningful protections around AI and wages that reflect the value of their labor," Crabtree-Ireland said9.
By the numbers
- 12.0 percent: Claude Opus 5.5's score on SemiAnalysis's new GDPx test, the best of 13 models graded1.
- $2.6 billion: Aramco's cash contribution for a 49 percent stake in HUMAIN4.
- Eleven billion dollars: what OpenAI's newest funding round has secured, against an original $30 billion target3.
- Zero: the score finance and legal tasks returned for every model tested on GDPx1.
- 11.1 thousand: GitHub stars DeepSeek's dsh-superagent had collected by Monday10.
- 12.6 percent: the wage increase SAG-AFTRA's new animation contracts give voice actors9.
- 50: the Samsung executives Sam Altman met in Yongin on Saturday8.
What to watch
SemiAnalysis's next GDPx version is due in three to six months. Watch whether it moves the 12 percent score, or just moves the leaderboard. Kapoor's reliability metrics are the harder number to fix.
Keep an eye on OpenAI's family-office pitch too. A $20 billion target that still falls short tells investors more than a press release would.
Sources
- Max Kan, Andrew Megalaa, Dylan Patel, "GDPx: Measuring Frontier Model Capabilities Across the Most Valuable Domains," SemiAnalysis, Nov. 2, 2026, https://newsletter.semianalysis.com/p/gdpx-measuring-frontier-model-capabilities
- Stephan Rabanser, Sayash Kapoor, Peter Kirgis, Kangheng Liu, Saiteja Utpala, Arvind Narayanan, "Towards a Science of AI Agent Reliability," arXiv, Feb. 18, 2026, https://arxiv.org/abs/2602.16666
- "OpenAI reportedly lowers funding target for new round," TechCrunch, Oct. 31, 2026, https://techcrunch.com/2026/10/31/openai-reportedly-lowers-funding-target-for-new-round/
- "Saudi Aramco Acquires 49% Stake in HUMAIN with SAR 9.9 Billion Investment to Accelerate Saudi Arabia's AI Ambitions," Business Today Middle East, Nov. 1, 2026, https://businesstoday.me/tech-news/saudi-aramco-acquires-49-stake-in-humain-with-sar-9-9-billion-investment-to-accelerate-saudi-arabias-ai-ambitions/
- Anamarija Pogorelec, "DeepSeek open-sources Super-Agent, a coding agent that watches team chats for work," Help Net Security, Nov. 2, 2026, https://www.helpnetsecurity.com/2026/11/02/deepseek-dsh-superagent/
- "'We don't want to ship AI for AI's sake,' says Apple's John Ternus," The Next Web, Nov. 1, 2026, https://thenextweb.com/news/john-ternus-wired-apple-ai-for-ais-sake
- "Microsoft AI CEO Mustafa Suleyman warns against building conscious machines, calls for pause if signs emerge," People Matters, Nov. 2, 2026, https://me.peoplemattersglobal.com/news/ai-and-emerging-tech/microsoft-ai-ceo-mustafa-suleyman-warns-against-building-conscious-machines-calls-for-pause-if-signs-emerge-53114
- "Altman turns to Samsung, SK as AI expansion faces chip supply challenges," The Korea Herald, Nov. 2, 2026, https://www.koreaherald.com/article/10922884
- "SAG-AFTRA Members Ratify TV Animation Contracts," AWN, Oct. 30, 2026, https://www.awn.com/news/sag-aftra-members-ratify-tv-animation-contracts
- "deepseek-ai/dsh-superagent-prototype," GitHub, Nov. 2, 2026, https://github.com/deepseek-ai/dsh-superagent-prototype
