Source: Epoch AI (CC BY 4.0). ECI is Epoch's item-response model over many benchmarks; each model page shows its 90% confidence interval.
Only models with both an ECI score and an OpenRouter price. Normalisation: 75% input + 25% output price. Click a point to open the model.
| # | Model | ECI ↓ | Context | In $/M | Weights | Released |
|---|---|---|---|---|---|---|
| 1 | Phi 4 | 130.4 | 16K | $0.070 | Open source | 12 Dec 2024 |
| 2 | phi-3-small 7.4B | 121.8 | — | — | Open source | 23 Apr 2024 |
| 3 | phi-3-medium 14B | 121.2 | — | — | Open source | 23 Apr 2024 |
| 4 | phi-3-mini 3.8B | 117.2 | — | — | Open source | 23 Apr 2024 |
| 5 | Phi-2 | 107.6 | — | — | Open source | 12 Dec 2023 |
| 6 | Phi-1.5 | 90.8 | — | — | Open source | 11 Sep 2023 |
| 7 | Phi-3.5-MoE | — | — | — | Open source | 17 Aug 2024 |
| 8 | Phi-3.5-mini | — | — | — | Open source | 16 Aug 2024 |
| 9 | WizardLM-2 8x22B | — | 66K | $0.62 | Open source | 16 Apr 2024 |