OpenAI Claims Math Breakthroughs, Gemini 4 Carbon Tops, GPT-6.1 Terra Cuts Costs
The week's biggest claim came from OpenAI: on 27 October its internal AI announced three research breakthroughs — a proof of local smoothness implying the 35-year-old Kakeya conjecture, plus a claimed refutation of the Exponential Time Hypothesis held for about 25 years. The proofs passed Lean formal verification and are public, awaiting academic scrutiny. In models, Google shipped Gemini 4 Carbon just two weeks after Argon, topping the Artificial Analysis Intelligence Index v5 with 59 points, focused on cybersecurity and long-horizon coding. OpenAI countered with GPT-6.1 Terra, a low-cost computer-use model priced at $1/$5 per million tokens, cutting per-task cost to $0.40 — roughly a ninth of Claude Opus 5.5. On capital, Meta filed to raise $95 billion for AI datacenters while Anthropic submitted its S-1 for a Nasdaq IPO, revealing $16.3 billion in H1 2026 revenue.
Five things this week
Enough to retellUS vs China
The same week, both sides| Dimension | United States | China | What it means for ASEAN |
|---|---|---|---|
| New models | Google launched Gemini 4 Carbon, topping AA v5; OpenAI shipped computer-use model GPT-6.1 Terra | Xiaomi open-sourced 0.8B document model Xiaomi-OCR-0; OpenBMB released open-source MiniCPM-OCR suite | Small open models suit localized, on-premise deployments |
| Prices | GPT-6.1 Terra costs $1/$5 per million tokens; per-task cost about one-ninth of Opus 5.5 | GLM 5.3 reproduced a Firefox exploit chain at roughly $1,000 per run in API fees | Agent task costs falling fast, good for SEA exporters |
| Benchmarks | Gemini 4 Carbon led security evals like SEC-bench Pro, first in 4 of 9 assessments | Seedance 2.5 topped AA image-to-video benchmark v2.1; GLM 5.3 passed Epoch AI safety testing | Security and video generation are the new arenas |
| Policy | Anthropic filed S-1 for Nasdaq listing; Meta declared $95 billion raise for datacenters | Alibaba Cloud open-sourced AI Agent Handbook; survey shows only ~18% agents in production | Disclosure tightening; useful compliance reference for Southeast Asia |