Labs & Countries

Labs & Countries is a data page on the AI desk of Glassbox Auditing. Which labs and countries build the best AI models, their gap to the leader now and a year ago, open against closed models, and what prediction-market bettors expect. The leading labs are the biggest customers for chips and data centers, and the U.S.-China gap shapes export-control policy.

Last updated: 7 Nov 2026.

What is the latest reading?

U.S. share of notable AI models: 59.6% as of 2026, +59.6 pts since 1969. Source: Epoch AI notable models catalogue (CC BY 4.0).

What is on this page today?

Which company has best AI model end of 2026: the betting crowd's favourite is Google at 64%, +11 points in a week — OpenAI leads Epoch's capabilities index, at 4% in this market · odds stored since 5 Jun 2026 · market ends 31 Dec 2026 · settles on the leaderboard its rules name (Polymarket, 7 Nov 2026) GOOGL
Which company has the best AI model end of November: the betting crowd's favourite is Google at 76%, +6 points in a week — OpenAI leads Epoch's capabilities index, at 3% in this market · odds stored since 19 Sep 2026 · market ends 1 Dec 2026 · settles on the leaderboard its rules name (Polymarket, 7 Nov 2026) GOOGL
Which company has the best AI model on LiveBench (Mathematics) end of November: the betting crowd's favourite is OpenAI at 1%, -1 point in a week — OpenAI leads Epoch's capabilities index, at 1% in this market · odds stored since 1 Oct 2026 · market ends 1 Dec 2026 · settles on the leaderboard its rules name (Polymarket, 7 Nov 2026)
Which company has the best AI model on LiveBench (Overall) end of November: the betting crowd's favourite is Anthropic at 62%, -6 points in a week — OpenAI leads Epoch's capabilities index, at 31% in this market · odds stored since 1 Oct 2026 · market ends 1 Dec 2026 · settles on the leaderboard its rules name (Polymarket, 7 Nov 2026)
Which company has the best AI model on LiveBench (Coding) end of December: the betting crowd's favourite is Anthropic at 69%, +12 points in a week — OpenAI leads Epoch's capabilities index, at 15% in this market · odds stored since 1 Nov 2026 · market ends 1 Jan 2027 · settles on the leaderboard its rules name (Polymarket, 7 Nov 2026)
Google has the model LMArena's voters rate highest, Gemini 4 Carbon (High) at 1539; Epoch's index ranks the lab #3 — best model per lab on LMArena's text leaderboard (blind votes), against the lab's rank on Epoch's capabilities index; different methods, so a lab placed differently is good at different things (LMArena, 5 Nov 2026)
France is the closest to the U.S. after China: its best model (Mistral Large 4) is 14.4 points behind, 18.3 a year earlier — points behind the best U.S. model on Epoch's capabilities index · about 11 months behind · United Kingdom, United Arab Emirates follow (Epoch AI Capabilities Index, 28 Oct 2026)
China's best AI model trails the U.S. best by 9.8 points (about 8.1 months behind), 5.6 a year earlier — DeepSeek V4.1 Flash 1017 against GPT-6 Astra (Epoch AI Capabilities Index, 16 Oct 2026)
DeepSeek ranks #5 on the capabilities index, 9.8 points behind the leader: DeepSeek V4.1 Flash 1017 at 159.3 — China · open weights · +14.9 points in 12 months · 7 models in 12 months (Epoch AI Capabilities Index, 16 Oct 2026)
Meta ranks #4 on the capabilities index, 7.0 points behind the leader: Muse Wonder at 162.0 — United States · +29.9 points in 12 months · 6 models in 12 months (Epoch AI Capabilities Index, 11 Oct 2026) META
Google ranks #3 on the capabilities index, 2.6 points behind the leader: Gemini 4 Argon at 166.4 — United States · +21.7 points in 12 months · 12 models in 12 months (Epoch AI Capabilities Index, 29 Sep 2026) GOOGL
The eight AI labs' companies had 33,926 H-1B petitions approved in fiscal 2026, -14% on fiscal 2025; Amazon 16,878 — approvals of new and continuing petitions across OpenAI, Anthropic, Google, Meta, Microsoft, NVIDIA, xAI, Amazon · small labs file few petitions, so their numbers are small but real · fiscal 2027 is still being counted (USCIS, 29 Sep 2026)
Anthropic ranks #2 on the capabilities index, 0.7 points behind the leader: Claude Opus 5.5 at 168.4 — United States · +20.8 points in 12 months · 12 models in 12 months (Epoch AI Capabilities Index, 21 Sep 2026)
OpenAI leads the capabilities index: GPT-6 Astra at 169.1 — United States · +17.9 points in 12 months · 17 models in 12 months (Epoch AI Capabilities Index, 2 Sep 2026)
Developers in the United States built 75 notable AI models in the last 12 months, China 35 — 60 in the 12 months before · models Epoch counts as notable (state of the art, highly cited or widely used), by the developers' country (Epoch AI notable models, 2 Sep 2026)
U.S. developers built 60% of notable AI models in 2025 and China's 33% (U.S. 59%, China 32% in 2024) — 109 notable models in 2025 · a model built across two countries counts for both, so the shares can add to more than 100% · the catalogue fills in over 12-18 months (Epoch AI notable models, 30 Dec 2025)

Where does the data come from?

Items on this page come from Polymarket; LMArena; Epoch AI Capabilities Index; USCIS; Epoch AI notable models. Each item links to its own source. All sources, how items are checked.

What does this page lead to?

Power Board (World Powers): The Power Board's AI column is the same comparison of each country's best model, so a shift in the lab rankings shows there. (moves the same way; same day; global scale) Source: Epoch AI Capabilities Index (CC BY 4.0).
Sanctions & Watchlists (Global Intelligence): The U.S.-China model gap is the case made for chip export controls; Commerce's Entity List, part of the Consolidated Screening List, names Chinese chip and AI firms U.S. suppliers need a license to ship to. (either way, depending on the cause; months; global scale) Source: US Consolidated Screening List (Commerce, State, Treasury).
AI Companies (AI): Each lab's best score is the capability column on its developer's profile, beside its funding and compute. (moves the same way; same day; company scale) Source: Epoch AI Capabilities Index (CC BY 4.0).
Polymarket (Markets): Polymarket's AI-ranking markets pay out on which company tops the LMArena leaderboard by a date, so a lab's rise or fall in the rankings moves the odds bettors give it. (either way, depending on the cause; same day; sector scale) Source: LMArena leaderboard dataset.
Business & Industry (AI): A small gap between open-weights and closed models lets firms run capable models on their own hardware or cheap hosts, which tends to lower what near-frontier AI costs a business. (either way, depending on the cause; months; sector scale) Source: OpenRouter model prices.

What feeds this page?

Benchmarks (AI): Best scores decide which lab and which country hold the lead; a new top result widens or narrows the gaps the Labs page measures, depending on who posts it. (either way, depending on the cause; same day; sector scale) Source: Epoch AI Benchmarking Hub (CC BY 4.0).
AI Researchers (AI): Frontier models come out of labs with deep research benches, so where leading authors work (OpenAlex records their institutions and countries) underlies the country gap the Labs page measures. (either way, depending on the cause; years; global scale) Source: OpenAlex (scholarly works).
Polymarket (Markets): Polymarket's AI ranking markets price which lab will hold the top model; Labs & Countries sets those odds beside measured benchmark results. (either way, depending on the cause; same day; sector scale) Source: Polymarket (public blockchain trades and wallet profiles).