Best score reached on each track, held flat until something beats it. The horizontal distance between the two staircases is time: how long the best downloadable model takes to reach a level hosted models already passed.
65 of 65 scored models, one row per model line
| # | Model | ECI |
|---|---|---|
| 1 | Proprietary model: Claude Fable 5Anthropic · Proprietary model | 163.2 |
| 2 | Proprietary model: Claude Opus 5Anthropic · Proprietary model | 162.4 |
| 3 | Proprietary model: GPT-5.5 ProOpenAI · Proprietary model | 162.1 |
| 4 | Open weights: Kimi K3Moonshot · Open weights | 157.7 |
| 5 | Proprietary model: Gemini 3.7 FlashGoogle DeepMind · Proprietary model | 157.3 |
| 6 | Proprietary model: Qwen 3.8 MaxAlibaba · Proprietary model | 156.6 |
| 7 | Proprietary model: Grok 4.6xAI · Proprietary model | 156.3 |
| 8 | Proprietary model: Claude Sonnet 5Anthropic · Proprietary model | 156.2 |
| 9 | Proprietary model: Muse Spark 1.2Meta AI · Proprietary model | 155.5 |
| 10 | Open weights: DeepSeek V4 Pro 0813DeepSeek · Open weights | 155.4 |
| 11 | Proprietary model: GLM-5.3Z.ai (Zhipu AI) · Proprietary model | 155.3 |
| 12 | Open weights: Kimi K2.7 CodeMoonshot · Open weights | 150.3 |
| 13 | Open weights: Inkling-SmallThinking Machines · Open weights | 150.2 |
| 14 | Proprietary model: Grok 4.3 BetaxAI · Proprietary model | 149.2 |
| 15 | Open weights: InklingThinking Machines · Open weights | 148.7 |
| 16 | Proprietary model: Qwen 3.6 PlusAlibaba · Proprietary model | 147.6 |
| 17 | Proprietary model: o3-proOpenAI · Proprietary model | 147.5 |
| 18 | Proprietary model: o3OpenAI · Proprietary model | 146.9 |
| 19 | Open weights: MiniMax-M3MiniMax · Open weights | 146.6 |
| 20 | Open weights: Nemotron 3 UltraNVIDIA · Open weights | 146.3 |
| 21 | Proprietary model: GPT-5.4 NanoOpenAI · Proprietary model | 145.9 |
| 22 | Proprietary model: o4-miniOpenAI · Proprietary model | 145.7 |
| 23 | Proprietary model: Gemini 3.5 Flash-LiteGoogle DeepMind · Proprietary model | 145.1 |
| 24 | Proprietary model: Grok 4 FastxAI · Proprietary model | 144.2 |
| 25 | Open weights: Gemma 4 31B ITGoogle DeepMind · Open weights | 142.7 |
Scores: Epoch AI Capabilities Index (CC BY), retrieved Sep 1, 2026
From the blog
Market Analysis: Open Weights vs Proprietary Models
How far behind the downloadable models really are, and who has been closing the gap.
Get the latest AI insights delivered to your inbox. No spam, unsubscribe anytime.