Road to AGI
Data as of

The frontier AI race · Public evidence

West leads the measured frontier. China leads the open frontier.

The path to AGI is now an ecosystem competition—not a single-company story. We track the strongest Western and Chinese systems, who is closing the gap, and where open weights change the balance.

See all tracked models What this does—and does not—measure

Current composite frontier

Western ecosystem GPT-5.6 Sol (pro, max) OpenAI
161.7
Chinese ecosystem Kimi K3 (max) Moonshot AI
155.6

Western lead: 6.1 ECI

Composite frontier West +6.1 Highest ECI by ecosystem
Open-weight frontier China +9.5 GLM-5.2
Labs above 145 ECI 5 — 5 Western — Chinese
Selected benchmark leads 3 — 1 1 without comparison

ECI frontier history

The gap is real. The direction is not settled.

Western ecosystem Chinese ecosystem

Highest dated Epoch Capabilities Index score in each ecosystem at year end. The comparison begins when both ecosystems have a scored model.

1101301501652023: GPT-4 (Mar 2023) — 125.91 ECI2023: Yi-34B — 117.12 ECI20232024: o1 (high) — 142.71 ECI2024: DeepSeek-V3 — 133 ECI20242025: GPT-5.2 Pro — 154.66 ECI2025: DeepSeek-V3.2 (Thinking; Fireworks) — 146.59 ECI20252026: GPT-5.6 Sol (pro, max) — 161.72 ECI2026: Kimi K3 (max) — 155.59 ECI2026West161.7China155.6
ECI frontier history by ecosystem. 2023: Western leader GPT-4 (Mar 2023), 125.91 ECI; Chinese leader Yi-34B, 117.12 ECI.2024: Western leader o1 (high), 142.71 ECI; Chinese leader DeepSeek-V3, 133 ECI.2025: Western leader GPT-5.2 Pro, 154.66 ECI; Chinese leader DeepSeek-V3.2 (Thinking; Fireworks), 146.59 ECI.2026: Western leader GPT-5.6 Sol (pro, max), 161.72 ECI; Chinese leader Kimi K3 (max), 155.59 ECI.
ECI combines 50+ benchmarks into a general-capability scale. It is not a percentage and does not estimate AGI arrival.

Two plausible paths

The winner depends on what kind of race this becomes.

01 / Closed frontier

If scale and proprietary systems decide it…

The Western ecosystem holds today’s measured edge, with OpenAI and Anthropic occupying the top three deduplicated ECI positions.

GPT-5.6 Sol (pro, max) · 161.7 ECI
02 / Open diffusion

If replication and open weights decide it…

The Chinese ecosystem has the stronger open-weight frontier, led by Zhipu, Moonshot, DeepSeek, Alibaba, and MiniMax.

GLM-5.2 · 151.4 ECI

Lab depth

This is no longer a two-company race.

Each bar shows a lab’s highest deduplicated ECI score. Depth matters: both ecosystems now have five tracked labs above 145 ECI.

Western ecosystem

  1. OpenAI 161.7
    GPT-5.6 Sol (pro, max) 3 open-weight
  2. Anthropic 160.7
    Claude Fable 5 (max)
  3. Google DeepMind 154.9
    Gemini 3.1 Pro Preview 8 open-weight
  4. Meta AI 154.3
    Muse Spark 22 open-weight
  5. xAI 153.6
    Grok 4.5 (high)
  6. Mistral AI 135.3
    Mistral Medium 3 12 open-weight
  7. Microsoft 131.1
    Phi-4 9 open-weight

Chinese ecosystem

  1. Moonshot AI 155.6
    Kimi K3 (max) 5 open-weight
  2. Alibaba / Qwen 153.1
    Qwen3.7-Max 26 open-weight
  3. Zhipu AI / Z.ai 151.4
    GLM-5.2 6 open-weight
  4. DeepSeek 148.7
    DeepSeek v4 Pro (max) 19 open-weight
  5. MiniMax 147.1
    MiniMax-M2.5 5 open-weight
  6. 01.AI 117.1
    Yi-34B 3 open-weight
  7. Baichuan 102.9
    Baichuan2-13B 4 open-weight
Explore the full model roster