How far behind the US is China in AI?

China's best AI model, Kimi K3.1 Pro, scores 46 on Artificial Analysis's index, a level the US first reached about 2 months earlier. The top US model with a public API, Claude Opus 5.5, scores 58 (October 2026).

2 months

China's Kimi K3.1 Pro (Moonshot AI) scores what the US first reached in September 2026. The leading US model is Claude Opus 5.5.

As of

Test by test

United StatesChina

TestScore out of 100US first got there
Hard computer tasksReal software tasks in a command lineTerminal-Bench 4.0US: Claude Opus 5.5 · China: GLM-5.3
65
42
Jun 20265 months agoClaude Fable 5
Office work across appsAutomationBenchUS: Gemini 4 Argon · China: Kimi K3
51
23
Jul 20264 months agoGPT-5.6 Sol
Real coding jobsDeepSWEUS: GPT-6 Astra · China: GLM-5.3
74
69
Jun 20265 months agoClaude Fable 5
Hardest exam questionsExpert questions across many subjectsHumanity's Last ExamUS: Claude Opus 5.5 · China: Kimi K3.1 Pro
61
52
Jun 20265 months agoClaude Fable 5
Long jobs in real work softwareAgents' Last ExamUS: Claude Opus 5.5 · China: Kimi K3
43
35
Sep 20262 months agoGPT-6 Astra

One more test uses another scale: Work tasks from 44 jobs, US got there Sep 2026 (1 month).

How close is AI to replacing programmers? →

The US has led in AI every month since 2023

Top model from each country by release date, on Epoch AI's capability score, which combines many tests into one scale.

  • United States
  • China
80100120140160180Jul '23Jan '24Jul '24Jan '25Jul '25Jan '26Jul '26Today, 1 Nov1 Jun 2023. China 90.6. Baichuan1-7B6 Sep 2023. China 103.3. Baichuan2-13B24 Sep 2023. China 113.1. Qwen-14B2 Nov 2023. China 117.4. Yi-34B7 May 2024. China 124.8. DeepSeek-V2 (MoE-236B, May 2024)7 Jun 2024. China 125.3. Qwen2-72B17 Sep 2024. China 128.5. Qwen2.5-32B19 Sep 2024. China 129. Qwen2.5-72B26 Dec 2024. China 132.3. DeepSeek-V320 Jan 2025. China 139. DeepSeek-R128 Apr 2025. China 139.4. Qwen3-235B-A22B28 May 2025. China 141.3. DeepSeek-R1 (May 2025)25 Jul 2025. China 143.8. Qwen3-235B-A22B-Thinking (Jul 2025)29 Sep 2025. China 145. DeepSeek-V3.2-Exp6 Nov 2025. China 146. Kimi K2 Thinking1 Dec 2025. China 146.1. DeepSeek-V3.227 Jan 2026. China 148. Kimi K2.57 Apr 2026. China 150. GLM-5.120 Apr 2026. China 150.9. Kimi K2.619 May 2026. China 153.8. Qwen3.7-Max16 Jul 2026. China 157.4. Kimi K324 Feb 2023. United States 110.2. LLaMA-65B14 Mar 2023. United States 125.9. GPT-4 (Mar 2023)25 Jan 2024. United States 126.5. GPT-4 Turbo (Nov 2023)29 Feb 2024. United States 126.9. Claude 3 Opus9 Apr 2024. United States 127.3. GPT-4 Turbo (Apr 2024)13 May 2024. United States 129.1. GPT-4o (May 2024)20 Jun 2024. United States 130. Claude 3.5 Sonnet12 Sep 2024. United States 135.8. o1-mini17 Dec 2024. United States 141.8. o131 Mar 2025. United States 144.1. Gemini 2.5 Pro (Mar 2025)16 Apr 2025. United States 146.9. o310 Jun 2025. United States 147.4. o3-pro7 Aug 2025. United States 150. GPT-57 Oct 2025. United States 150.7. GPT-5 Pro18 Nov 2025. United States 152.9. Gemini 3 Pro11 Dec 2025. United States 156. GPT-5.2 Pro5 Feb 2026. United States 156.8. GPT-5.3 Codex5 Mar 2026. United States 159. GPT-5.4 Pro23 Apr 2026. United States 162. GPT-5.5 Pro9 Jun 2026. United States 162.1. Claude Fable 524 Jul 2026. United States 162.6. Claude Opus 51 Sep 2026. United States 164.7. Claude Fable 5.13 Sep 2026. United States 166.4. GPT-6 Astra22 Sep 2026. United States 166.8. Claude Opus 5.5United States166.8China157.480100120140160180Jan '24Jan '25Jan '26Today, 1 Nov1 Jun 2023. China 90.6. Baichuan1-7B6 Sep 2023. China 103.3. Baichuan2-13B24 Sep 2023. China 113.1. Qwen-14B2 Nov 2023. China 117.4. Yi-34B7 May 2024. China 124.8. DeepSeek-V2 (MoE-236B, May 2024)7 Jun 2024. China 125.3. Qwen2-72B17 Sep 2024. China 128.5. Qwen2.5-32B19 Sep 2024. China 129. Qwen2.5-72B26 Dec 2024. China 132.3. DeepSeek-V320 Jan 2025. China 139. DeepSeek-R128 Apr 2025. China 139.4. Qwen3-235B-A22B28 May 2025. China 141.3. DeepSeek-R1 (May 2025)25 Jul 2025. China 143.8. Qwen3-235B-A22B-Thinking (Jul 2025)29 Sep 2025. China 145. DeepSeek-V3.2-Exp6 Nov 2025. China 146. Kimi K2 Thinking1 Dec 2025. China 146.1. DeepSeek-V3.227 Jan 2026. China 148. Kimi K2.57 Apr 2026. China 150. GLM-5.120 Apr 2026. China 150.9. Kimi K2.619 May 2026. China 153.8. Qwen3.7-Max16 Jul 2026. China 157.4. Kimi K324 Feb 2023. United States 110.2. LLaMA-65B14 Mar 2023. United States 125.9. GPT-4 (Mar 2023)25 Jan 2024. United States 126.5. GPT-4 Turbo (Nov 2023)29 Feb 2024. United States 126.9. Claude 3 Opus9 Apr 2024. United States 127.3. GPT-4 Turbo (Apr 2024)13 May 2024. United States 129.1. GPT-4o (May 2024)20 Jun 2024. United States 130. Claude 3.5 Sonnet12 Sep 2024. United States 135.8. o1-mini17 Dec 2024. United States 141.8. o131 Mar 2025. United States 144.1. Gemini 2.5 Pro (Mar 2025)16 Apr 2025. United States 146.9. o310 Jun 2025. United States 147.4. o3-pro7 Aug 2025. United States 150. GPT-57 Oct 2025. United States 150.7. GPT-5 Pro18 Nov 2025. United States 152.9. Gemini 3 Pro11 Dec 2025. United States 156. GPT-5.2 Pro5 Feb 2026. United States 156.8. GPT-5.3 Codex5 Mar 2026. United States 159. GPT-5.4 Pro23 Apr 2026. United States 162. GPT-5.5 Pro9 Jun 2026. United States 162.1. Claude Fable 524 Jul 2026. United States 162.6. Claude Opus 51 Sep 2026. United States 164.7. Claude Fable 5.13 Sep 2026. United States 166.4. GPT-6 Astra22 Sep 2026. United States 166.8. Claude Opus 5.5United States166.8China157.4100125150175Jan '24Jan '25Jan '26Today, 1 Nov1 Jun 2023. China 90.6. Baichuan1-7B6 Sep 2023. China 103.3. Baichuan2-13B24 Sep 2023. China 113.1. Qwen-14B2 Nov 2023. China 117.4. Yi-34B7 May 2024. China 124.8. DeepSeek-V2 (MoE-236B, May 2024)7 Jun 2024. China 125.3. Qwen2-72B17 Sep 2024. China 128.5. Qwen2.5-32B19 Sep 2024. China 129. Qwen2.5-72B26 Dec 2024. China 132.3. DeepSeek-V320 Jan 2025. China 139. DeepSeek-R128 Apr 2025. China 139.4. Qwen3-235B-A22B28 May 2025. China 141.3. DeepSeek-R1 (May 2025)25 Jul 2025. China 143.8. Qwen3-235B-A22B-Thinking (Jul 2025)29 Sep 2025. China 145. DeepSeek-V3.2-Exp6 Nov 2025. China 146. Kimi K2 Thinking1 Dec 2025. China 146.1. DeepSeek-V3.227 Jan 2026. China 148. Kimi K2.57 Apr 2026. China 150. GLM-5.120 Apr 2026. China 150.9. Kimi K2.619 May 2026. China 153.8. Qwen3.7-Max16 Jul 2026. China 157.4. Kimi K324 Feb 2023. United States 110.2. LLaMA-65B14 Mar 2023. United States 125.9. GPT-4 (Mar 2023)25 Jan 2024. United States 126.5. GPT-4 Turbo (Nov 2023)29 Feb 2024. United States 126.9. Claude 3 Opus9 Apr 2024. United States 127.3. GPT-4 Turbo (Apr 2024)13 May 2024. United States 129.1. GPT-4o (May 2024)20 Jun 2024. United States 130. Claude 3.5 Sonnet12 Sep 2024. United States 135.8. o1-mini17 Dec 2024. United States 141.8. o131 Mar 2025. United States 144.1. Gemini 2.5 Pro (Mar 2025)16 Apr 2025. United States 146.9. o310 Jun 2025. United States 147.4. o3-pro7 Aug 2025. United States 150. GPT-57 Oct 2025. United States 150.7. GPT-5 Pro18 Nov 2025. United States 152.9. Gemini 3 Pro11 Dec 2025. United States 156. GPT-5.2 Pro5 Feb 2026. United States 156.8. GPT-5.3 Codex5 Mar 2026. United States 159. GPT-5.4 Pro23 Apr 2026. United States 162. GPT-5.5 Pro9 Jun 2026. United States 162.1. Claude Fable 524 Jul 2026. United States 162.6. Claude Opus 51 Sep 2026. United States 164.7. Claude Fable 5.13 Sep 2026. United States 166.4. GPT-6 Astra22 Sep 2026. United States 166.8. Claude Opus 5.5United States166.8China157.4

Distillation attacks on American AI

Several Chinese labs have trained their own models on Claude's answers. Anthropic and US agencies measured how much.

Exchanges with Claude that Anthropic traced to each lab, May to July 2026

Anthropic, Sep 2026

Almost 300,000

customer requests Moonshot sent to Claude in 10 days, serving Claude's replies to its own users

Anthropic, Sep 2026

6 labs

named by the NSA, CISA and FBI for copying Claude, GPT, Gemini and Grok: DeepSeek, Moonshot, Alibaba, MiniMax, StepFun and Z.AI

US advisory AA26-251A, Sep 2026

90–99%

less data collected by copiers since Anthropic's new defences

Anthropic, Oct 2026

See how AI is used in hacking →

Sources (11)
  1. Artificial Analysis, artificialanalysis.ai/leaderboards/models
  2. Terminal-Bench, tbench.ai/leaderboard
  3. Zapier, zapier.com/benchmarks
  4. Datacurve, deepswe.datacurve.ai
  5. Artificial Analysis, artificialanalysis.ai/evaluations/humanitys-last-exam
  6. UC Berkeley RDI, agents-last-exam.org/leaderboard
  7. Artificial Analysis, artificialanalysis.ai/evaluations/gdpval-aa
  8. Epoch AI, Epoch Capabilities Index
  9. Anthropic, Chinese-speaking operators used Claude on about 50 organisations, Sep 2026
  10. CISA, cisa.gov/news-events/cybersecurity-advisories/aa26-251a
  11. Anthropic, anthropic.com/research/distillation-defenses