whatmodel.sh / Phi-3-vision-128k-instruct
Phi-3-vision-128k-instruct
↗ share
+ contribute
Phi-3-vision-128k-instruct is an open-weights language model from Microsoft, released May 2024. Its published size is 4.15B parameters. Licence: mit.
Catalogue identity notes
Comparison-only. Arena entry 'phi-3-vision-128k-instruct' (vision). LMArena's leaderboard table (arena-leaderboard space @6f2eeeb) labels 'phi-3-vision-128k-instruct' as 'Phi-3-Vision-128K-Instruct' (Microsoft, MIT), linking https://huggingface.co/microsoft/Phi-3-vision-128k-instruct. Maker repo microsoft/Phi-3-vision-128k-instruct @ed2772f: licence tag mit. 4,146,621,440 parameters from the Hugging Face safetensors metadata (dense, so active = total). Released 2024-05: card: 'The model weight is released on May 21, 2024'.
$ how it compares
The LLMs ranked either side of Phi-3-vision-128k-instruct, out of 828 with a score, then more from Microsoft.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 600 |  WizardCoder-15B-V1.0 | 42±1 | local | ||
| 601 |  Snowflake Arctic Instruct | 41±2 | local | ||
| 602 | Phi-3-vision-128k-instruct | 41±2 | local | ||
| 603 |  OpenChat 3.5 | 41±2 | local | ||
| 604 |  InternVL2-4B | 41±2 | local | ||
| more from Microsoft | |||||
| 252 | Phi-4-reasoning-plus 14B | 59±1 | local | ||
| 261 | Phi-4-reasoning 14B | 58±1 | api $0.22/M +local | ||
| 365 | Phi-4 14B | 53±1 | api $0.087/M +local | ||
$ what it's good at
math▁▂▃▅▆▅▄█▄▁▁▁ #303 of 398 * · 6.3% of its score · 1 test
Beats GPT-4 (0613), Athene-70B, Llama 3 70B Instruct here, though they rank higher overall. Trails Gemma 7B (base), DeepSeek R1 Distill Qwen 1.5B, CodeGemma 7B (v1.0) here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| MathVista (testmini) lab | 44.5% ✎ | ▁··▁▄▄▇█▅▆█▄ beats 4/47 | 6.3% |
vision & documents▃▁▃▂▁▃▄▂▆▇█▄ #147 of 147 · 99% of its score · 9 tests
Trails Qwen2-VL 2B Instruct, Phi-3.5-vision-instruct, Qwen3.5 0.8B here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| CharXiv descriptive questions | 60.5% ✎ | ▄█▄▄▄·██·▄·█ beats 3/12 | 41% |
| Arena vision (style controlled) | 881.1 ✎ | ▃▁▂▂▂▂▄▅▅▆██ beats 0/145 | 20% |
| CharXiv (reasoning questions) | 31.6% ✎ | ▂▂▂█▃▂·▂▂··▂ beats 4/13 | 15% |
| MMMU (val) lab | 40.4% ✎ | ▁····▁▃▄█▇▆▆ beats 2/52 | 10% |
| MathVista (testmini) lab | 44.5% ✎ | ▁··▁▄▄▇█▅▆█▄ beats 4/47 | 6.3% |
| PhysBench physical-world understanding (test, Avg) | 38.4% ✎ | ▅▃█··▅▃····▃ beats 2/9 | 3.4% |
| ChartQA lab | 81.4% ✎ | ▁··▁▁▁▁▁▂▅▅█ beats 16/44 | 1.3% |
| TextVQA lab | 70.9% ✎ | ▁▁▁▄▅▁▃·▃▃▃█ beats 9/24 | 1.2% |
$ where to run it
Other ways to run it
- Run it as published · needs ~8.29 GB, NVIDIA only · from github.com/vllm-project
vllm serve microsoft/Phi-3-vision-128k-instruct --trust-remote-code
Official weights and documentation: huggingface.co/microsoft ↗
$ how the score was worked out · CharXiv leaderboard (princeto… 56%, LMArena leaderboard dataset 20%, Phi-3 Technical Report (Micro… 19% and 1 more
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 CharXiv leaderboard (princeton-nlp/CharXiv README) | 2 | 56% |
 LMArena leaderboard dataset | 1 | 20% |
 Phi-3 Technical Report (Microsoft, arXiv 2404.14219v3) · lab-reported | 5 | 19% |
| PhysBench: Benchmarking and Enhancing Vision-Language Models for Physical World Understanding (arXiv:2501.16411v2) | 1 | 3% |
| prior · pull toward the middle | 1% | |
| weighted average | = 41.4 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json