Phi-2
Phi-2 (Microsoft) was released on Dec 12, 2023. In the Epoch AI capability ranking (ECI), it ranks #226 out of 267 models, with 108.02. Against models of the same generation, its best result is on BIG-Bench Hard: 59.4%, #5 of 27.
Where Phi-2 stands out and where it falls behind, by the numbers
Phi-2 has published results on 8 tests compiled by Epoch AI. The yardstick here is its generation: its position among the models released between Mar 17, 2023 and Mar 11, 2024 that took the same test. Against the whole archive, which includes 2023 models, a recent release lands near the top of almost everything, and that says nothing.
Where it stands out
Tests where it sits in the top third of its own generation.
Where it falls behind
Tests where it sits in the bottom half of its own generation.
See all 8 tests, one by one
| Test | Score | In its generation | Overall | Test record |
|---|---|---|---|---|
| MMLU | 58.4% | #24 / 48 | #61 / 92 | 88.1% GPT-4o (May 2024) |
| ARC (AI2) | 75.9% | #10 / 45 | #18 / 65 | 95.3% DeepSeek-V3 (Mar 2025) |
| WinoGrande | 54.7% | #36 / 38 | #59 / 61 | 89.2% Llama 3.1-405B |
| HellaSwag | 53.6% | #30 / 32 | #48 / 51 | 95.3% GPT-4 (Mar 2023) |
| BIG-Bench Hard | 59.4% | #5 / 27 | #14 / 41 | 89.2% Gemini 1.5 Pro (Sept 2024) |
| OpenBookQA | 73.6% | #5 / 24 | #9 / 32 | 88.0% phi-3-mini 3.8B |
| TriviaQA | 45.2% | #19 / 19 | #32 / 32 | 87.6% Llama 2-70B |
| Adversarial Nli | 42.5% | #5 / 5 | #9 / 9 | 58.1% GPT-3.5 Turbo (Nov 2023) |
For each test, the model's best effort level counts, as in the ranking. Source of the scores: Epoch AI, under CC BY 4.0.
What to compare Phi-2 with
The models from other labs closest in capability and released in the same period, at most two per lab. Click any of them to see its page.
| Model | Lab | ECI | ELO | LiveBench | Release |
|---|---|---|---|---|---|
| Falcon 2 11B | Technology Innovation Institute | 109.63 | n/a | n/a | May 9, 2024 |
| Mistral 7B v0.3 | Mistral AI | 109.06 | n/a | n/a | May 27, 2024 |
| PaLM 2-M | 108.37 | n/a | n/a | May 17, 2023 | |
| Phi-2 | Microsoft | 108.02 | n/a | n/a | Dec 12, 2023 |
| Nemotron-4 15B | Nvidia | 107.80 | n/a | n/a | Feb 26, 2024 |
| Yi-9B | 107.69 | n/a | n/a | Mar 1, 2024 | |
| Qwen-7B | Alibaba | 106.91 | n/a | n/a | Sep 28, 2023 |
Phi-2 or Nemotron-4 15B?
Nemotron-4 15B is the model from another lab closest to Phi-2 in capability. Across the 5 tests both took, this is the score:
| Test | Phi-2 | Nemotron-4 15B |
|---|---|---|
| HellaSwag | 53.6% | 82.4% |
| WinoGrande | 54.7% | 78.0% |
| ARC (AI2) | 75.9% | 55.5% |
| BIG-Bench Hard | 59.4% | 58.7% |
| MMLU | 58.4% | 58.7% |
Unweighted score: each test counts one point, whether math or coding. The dot marks who did better on each row.
Other models from Microsoft
The models from the same lab closest in time, newest first.
| Model | Release | ECI | ELO | LiveBench | Access |
|---|---|---|---|---|---|
| phi-3-small 7.4B | Apr 23, 2024 | 121.85 | n/a | n/a | Open weights (unrestricted) |
| phi-3-medium 14B | Apr 23, 2024 | 121.25 | n/a | n/a | Open weights (unrestricted) |
| phi-3-mini 3.8B | Apr 23, 2024 | 117.40 | n/a | n/a | Open weights (unrestricted) |
| Phi-2 | Dec 12, 2023 | 108.02 | n/a | n/a | Open weights (unrestricted) |
| Phi-1.5 | Sep 11, 2023 | 91.68 | n/a | n/a | Open weights (unrestricted) |
Have you used Phi-2? Rate it
Users' opinion, not measurement: the rating is kept apart from the ranking and does not move anyone's position. One vote per browser, no sign-up, and you can change your rating later.
What people ask about Phi-2
Where does Phi-2 rank among AI models?
In the ROO3 AI Benchmark, ordered by the Epoch AI ECI, Phi-2 ranks #226 out of 267 models, with an ECI of 108.02. The ECI is a composite index that combines several hard benchmarks into a single scale. The position changes when a new model comes out, and the page is refreshed six times a day.
Is Phi-2 better than Nemotron-4 15B?
Across the 5 tests both models took and Epoch AI compiled, Phi-2 comes out ahead on 2, behind on 3 and tied on 0. It is a test-by-test count with no weighting: a win in math counts the same as a win in coding. To choose between the two, look at the tests that resemble your own use.
What is Phi-2 best at?
Against models of the same generation (released up to 9 months before and up to 3 months after it), the best relative results of Phi-2 are: BIG-Bench Hard (#5 of 27); OpenBookQA (#5 of 24); ARC (AI2) (#10 of 45). The yardstick is its position among the models of that period that took the same test.
Where do the numbers on this page come from?
From the same three sources as the ranking, all openly licensed: Epoch AI (ECI and individual tests, CC BY 4.0), Arena (human-preference ELO, CC BY 4.0) and LiveBench (overall average, CC BY-SA 4.0). ROO3 does not measure any model. When the page shows a figure published by the maker itself, it appears in the written analysis, labeled as the maker's, and never enters the ranking.
Where each number comes from
Epoch AI, "AI Benchmarking Hub". Published at epoch.ai. Data under the CC BY 4.0 licence, reordered by roo3.co.
The Arena (LMArena) leaderboard, from the public lmarena-ai/leaderboard-dataset on Hugging Face, under the CC BY 4.0 licence. We are not affiliated with the Arena.
LiveBench, by the LiveBench team, published at livebench.ai under the CC BY-SA 4.0 license. The overall average is recomputed by roo3.co with the same formula as the website (average of the categories), and those numbers stay under CC BY-SA 4.0. We are not affiliated with LiveBench.
← See the full ranking