[whatmodel.sh](../../README.md) / DeepSeek-V2.5
# DeepSeek-V2.5
DeepSeek · released 2024-09 · 236B params (21B active) · deepseek
**score 54.6** ±0.8 · #292 of 732 · from 13 benchmark results across 7 sources
Score within the [LMArena (text)](https://whatmodel.sh/?set=lmarena-text&score=set) comparison set only.
## What it's good at
● marks DeepSeek-V2.5 among every model on that task or test, worst to best.
task vs the others rank counts tests
coding ▁▁▁▃▅▇█●▅▄▃▁ #197 of 469 32% 4
chat & writing ▁▁▁▃▆▄▆●▆▇▇▃ #220 of 402* 9.2% 2
math ▁▁▃▅●▆▄█▄▂▁▁ #288 of 431 15% 3
coding · #197 of 469 · 32% of its score
Beats [Gemini 2.5 Flash](gemini-2-5-flash.md), [Qwen3 30B-A3B Instruct 2507](qwen3-30b-a3b-instruct-2507.md), [Qwen3.5 4B](qwen3-5-4b.md) here, though they rank higher overall. Trails [QwQ-32B-Preview](qwq-32b-preview.md), [Ministral 3 8B Reasoning 2512](ministral-3-8b-reasoning.md), [OpenThinker 32B](openthinker-32b.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
CRUXEval (ZeroEval, 800 problems) 70.0% ▁▂▆▃▆●▆▃▂·▁▁ beats 29/47 medium · avg 57% 14% sharp · at its level
Aider code editing (133 Exercism … 72.2% ▁·▁▂▄▁▆▄█●▂▂ beats 54/62 medium · avg 51% 11% sharp · at its level
WebApp1K-Duo-React pass@1 (onekq-… 49.0% ▁··▁·▁▁●▆█▇█ beats 6/45 hard · avg 29% 3.6% at its level
WebApp1K-React pass@1 (onekq-ai) 83.4% ▃▁▁█▁▄▅▃●▇▁▄ beats 22/31 medium · avg 40% 2.8% sharp · at its level · noisy
chat & writing · #220 of 402 · 9.2% of its score
Beats [Gemma 2 9B IT SimPO](gemma-2-9b-simpo.md), [Granite 4.2 8B](granite-4-2-8b.md), [Mistral Small 4 (119B-A6.5B)](mistral-small-4.md) here, though they rank higher overall. Trails [LFM2.5 1.2B Instruct](lfm2-5-1-2b-instruct.md), [Granite 4.0 H Micro 3B](granite-4-0-h-micro.md), [Gemma 3n E4B](gemma-3n-e4b.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
LMArena (text) 1307.2 ▁▁▁▂▃▃▃●▆▇█▃ beats 137/385 avg 1260 6.2% blunt
IFEval prompt-level strict 77.6% ●▄··▄·▄··█·▄ beats 0/6 easy · avg 73% 3.0% typical
math · #288 of 431 · 15% of its score
Beats [Molmo 2 8B](molmo-2-8b.md), [Llama 3.3 70B Instruct](llama-3-3-70b.md) here, though they rank higher overall. Trails [DeepSeek R1 Distill Qwen 1.5B](deepseek-r1-distill-qwen-1-5b.md), [Gemma 7B (base)](gemma-7b-base.md), [Mistral 7B v0.1 (base)](mistral-7b-v0-1.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
MATH Level 5 (ZeroEval, 721 probl… 44.7% ▁▁▁▆█▂●▇▄··▂ beats 22/38 hard · avg 30% 8.3% sharp · at its level
GSM8K (ZeroEval, zero-shot) 92.5% ▁▄·▁▄█▅▅▅●██ beats 26/43 very easy · avg 87% 4.6% sharp · maxed out
MATH-500 (maker) 74.8% ▁▃▂▃●▇▅▇█▄▂▁ beats 17/85 easy · avg 76% 2.0% small run
other tests (4)
test result vs the others in its run difficulty counts why
MMLU-Redux (ZeroEval, 2,778 quest… 80.3% ▁·▁▁▄▄▄●█▄▂▁ beats 29/53 easy · avg 75% 22% typical
ZebraLogic (ZeroEval, 1,000 puzzl… 22.1% ▁▁▃▆●█▁▁·▁▁▁ beats 34/57 hard · avg 15% 12% sharp · at its level
MMLU-Pro 65.8% ▃▂▅█▃·●▃▅··▂ beats 17/23 medium · avg 54% 10% at its level
DROP F1, 3-shot (simple-evals) 81.4% ▅▂▆●▆·▃·▂▅·▂ beats 9/23 easy · avg 81% 0.3% blunt · small run
source results share of the score
WildEval ZeroEval published results 5 61% ▇▇▇▇▇▇▇
Aider code editing benchmark runs 1 11% ▇
MMLU-Pro author-released evaluation runs 1 10% ▇
WebApp1K models leaderboard 2 6% ▇
LMArena leaderboard dataset 1 6% ▇
OpenCompass CompassAcademic leaderboard 2 3% ▇
DeepSeek vendor-reported benchmark scor… 1 2% ▇
prior: pull toward the middle 0% ▇
weighted average = 54.6
Licences: WildEval ZeroEval published results ([Apache-2.0](../../licenses/zeroeval-published-results-Apache-2.0.txt)) · Aider code editing benchmark runs ([Apache-2.0](../../licenses/aider-polyglot-runs-Apache-2.0.txt)) · MMLU-Pro author-released evaluation runs ([Apache-2.0](../../licenses/mmlu-pro-author-runs-Apache-2.0.txt)) · WebApp1K models leaderboard ([MIT](../../licenses/webapp1k-leaderboard-MIT.txt)) · LMArena leaderboard dataset ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · OpenCompass CompassAcademic leaderboard ([Apache-2.0](https://www.apache.org/licenses/LICENSE-2.0))
## How to run it
### llama.cpp (recommended)
1. Install llama.cpp. Homebrew and winget installs name the commands llama-cli and llama-server instead. [Other ways to install](https://github.com/ggml-org/llama.cpp/blob/master/docs/install.md)
```sh
# macOS / Linux
curl -LsSf https://llama.app/install.sh | sh
# Windows
irm https://llama.app/install.ps1 | iex
```
2. Run DeepSeek-V2.5 (Q4_K_M, needs ~142.5 GB, from [huggingface.co/lmstudio-community](https://huggingface.co/lmstudio-community/DeepSeek-V2.5-GGUF/tree/29ed5ced5bc86827a9ef071836b8de50434045ed))
```sh
llama cli -hf lmstudio-community/DeepSeek-V2.5-GGUF:Q4_K_M
```
3. Or start a local server and open the address it prints (http://localhost:9931)
```sh
llama serve -hf lmstudio-community/DeepSeek-V2.5-GGUF:Q4_K_M
```
Other versions: Q8_0 (~250.6 GB) `llama cli -hf lmstudio-community/DeepSeek-V2.5-GGUF:Q8_0`
### LM Studio
1. Install LM Studio. Open the app once before using the lms command. [Download it (macOS / Windows / Linux)](https://lmstudio.ai/download)
2. Run DeepSeek-V2.5 (Q4_K_M, needs ~142.5 GB, from [huggingface.co/lmstudio-community](https://huggingface.co/lmstudio-community/DeepSeek-V2.5-GGUF/tree/29ed5ced5bc86827a9ef071836b8de50434045ed))
```sh
lms get https://huggingface.co/lmstudio-community/DeepSeek-V2.5-GGUF@Q4_K_M -y
```
3. Then run lms ls to see its name, and lms chat followed by that name
Other versions: Q8_0 (~250.6 GB) `lms get https://huggingface.co/lmstudio-community/DeepSeek-V2.5-GGUF@Q8_0 -y`
### Ollama
1. Install Ollama. [Download it (macOS / Windows)](https://ollama.com/download)
```sh
# Linux
curl -fsSL https://ollama.com/install.sh | sh
```
2. Run DeepSeek-V2.5 (Q4_0, needs ~142.1 GB, from [ollama.com](https://ollama.com/library/deepseek-v2.5/tags))
```sh
ollama run deepseek-v2.5:236b-q4_0
```
Other versions: Q8_0 (~283.7 GB) `ollama run deepseek-v2.5:236b-q8_0`
Model repository and documentation: [huggingface.co/deepseek-ai](https://huggingface.co/deepseek-ai/DeepSeek-V2.5)
[interactive version](https://whatmodel.sh/model/deepseek-v2-5) · [JSON](https://whatmodel.sh/api?format=json&all=1)