[whatmodel.sh](../../README.md) / DeepSeek V4 Flash (Preview, Apr 2026)
# DeepSeek V4 Flash (Preview, Apr 2026)
DeepSeek · released 2026-04 · 284B params (13B active) · mit
**score 70.4** ±0.5 · #76 of 732 · from 58 benchmark results across 10 sources
Score within the [LMArena (text)](https://whatmodel.sh/?set=lmarena-text&score=set) comparison set only.
## What it's good at
● marks DeepSeek V4 Flash (Preview, Apr 2026) among every model on that task or test, worst to best.
task vs the others rank counts tests
agent tasks ▁▁▃▆█▄▅▅●▄▂▁ #70 of 313 24% 13
coding ▁▁▁▃▅▇█▇▅●▃▁ #77 of 469 23% 7
chat & writing ▁▁▁▃▆▄▆█▆▇●▃ #69 of 402* 2.9% 1
math ▁▁▃▅█▆▄█●▂▁▁ #51 of 431 27% 13
web search ▁▁●▅▄█▃▄·▁·▁ #74 of 84* 2.7% 2
agent tasks · #70 of 313 · 24% of its score
Beats [LongCat 2.0 (1.6T-A48B)](longcat-2-0.md), [GPT-5 Pro](gpt-5-pro.md), [GPT-5](gpt-5.md) here, though they rank higher overall. Trails [MiMo-V2.5 (310B-A15B)](mimo-v2-5.md), [GLM-5.1](glm-5-1.md), [GPT-5.4 mini](gpt-5-4-mini.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
SWE-bench Verified 73.5% █····●█·████ beats 1/6 hard · avg 21% 5.9% sharp · several runs
SWE-bench Multilingual 75.0% ▄··▄▄···▄●·█ beats 4/6 hard · avg 30% 2.5% at its level
Terminal-Bench 2.1 54.2% █··●█·▄····▄ beats 2/6 very hard · avg 13% 2.4% sharp · at its level
GameLogicBench solve rate (72 God… 16.7% ●▄·█·▄·██▄·▄ beats 0/11 very hard · avg 5.1% 1.7% at its level
Terminal-Bench Hard 34.1% ▄········█●▄ beats 4/5 very hard · avg 4.6% 1.7% sharp · at its level
Terminal-Bench 2.0 56.6% ▂▂▅▅███●▆▂▃▃ beats 24/36 very hard · avg 10% 1.6% at its level · small run
APEX-Agents (pass@1) 13.2% ▄▄····▄···●▄ beats 3/5 very hard · avg 1.9% 1.6% typical
τ³-bench 73.7% ▄····▄·█·▄·● beats 6/6 medium · avg 42% 1.6% at its level · several runs
Humanity's Last Exam with tools 45.1% ▄····●·█·▄·▄ beats 1/5 hard · avg 15% 1.5% at its level
BrowseComp 46.9% ▄·▄●···▄··█▄ beats 2/6 hard · avg 18% 1.2% at its level
MCP-Atlas (public set) 58.2% ▄▄▄·······●▄ beats 4/5 hard · avg 19% 0.9% at its level
τ³-bench telecom (tau2-bench ≥ v1… 98.3% ▄·▄···▄·▄·█● beats 6/6 easy · avg 79% 0.6% maxed out
τ³-bench Banking 22.3% █·▄▄······▄● beats 5/5 very hard · avg 5.3% 0.4% blunt
coding · #77 of 469 · 23% of its score
Beats [GPT-5.2](gpt-5-2.md), [GPT-5](gpt-5.md) here, though they rank higher overall. Trails [Llama 3.1 Nemotron Ultra 253B v1](llama-3-1-nemotron-ultra-253b.md), [MiMo-V2.5 (310B-A15B)](mimo-v2-5.md), [MiMo-V2.5-Pro (1T-A42B)](mimo-v2-5-pro.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
SWE-bench Verified 73.5% █····●█·████ beats 1/6 hard · avg 21% 5.9% sharp · several runs
Arena WebDev (Code Arena) 1430.3 ▁▁▃▂▅●▃▆▅▅▁▂ beats 52/129 avg 981 5.1% sharp
LiveCodeBench v6 (window unspecif… 90.9% ██···█··██●█ beats 5/6 medium · avg 46% 3.5% several runs
SWE-bench Multilingual 75.0% ▄··▄▄···▄●·█ beats 4/6 hard · avg 30% 2.5% at its level
Luau Leetcode fully solved (Torpe… 57.2% ▁▁▁▁·▁▂●▅▃█▇ beats 13/49 hard · avg 17% 2.4% sharp · at its level
GameLogicBench solve rate (72 God… 16.7% ●▄·█·▄·██▄·▄ beats 0/11 very hard · avg 5.1% 1.7% at its level
SciCode subproblems 48.2% ▃····▃··●·▃▃ beats 4/6 hard · avg 19% 1.5% at its level
chat & writing · #69 of 402 · 2.9% of its score
Beats [Claude Haiku 5.5](claude-haiku-5-5.md), [DeepSeek V4 Flash (0731)](deepseek-v4-flash.md), [GPT-5.2](gpt-5-2.md) here, though they rank higher overall. Trails [Nemotron Cascade 2 30B-A3B](nemotron-cascade-2-30b-a3b.md), [Gemma 4 31B](gemma-4-31b.md), [Gemini 2.5 Pro](gemini-2-5-pro.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
LMArena (text) 1438.4 ▁▁▁▂▃▃▃▆▆▇●▃ beats 301/385 avg 1260 2.9% blunt
math · #51 of 431 · 27% of its score
Beats [Grok 4.7](grok-4-7.md), [GLM-5.3-Flash](glm-5-3-flash.md), [LongCat 2.0 (1.6T-A48B)](longcat-2-0.md) here, though they rank higher overall. Trails [A.X K2 (688B-A33B)](ax-k2.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
MathArena ArXivMath April 2026 (4… 47.5% ▁·····▅●█▄·▁ beats 5/20 very hard · avg 7.6% 6.0% sharp · at its level
MathArena ArXivMath February 2026… 43.8% ▁▁▁▁█▇●▃▁▃·▃ beats 16/24 very hard · avg 9.4% 3.4% at its level
MathArena ArXivMath March 2026 (3… 52.5% █▃●·█▃▃·▅·▃▃ beats 5/14 hard · avg 16% 2.8% at its level
AIME 2026 97.0% ▄▄··▄····█·● beats 5/5 medium · avg 65% 2.4% sharp · maxed out · several runs
IMO-AnswerBench 91.1% ▄····▄·▄▄·●▄ beats 4/6 medium · avg 41% 2.4% several runs
MathArena ArXivMath May 2026 (40 … 55.8% ▁··▁▄●▄▁·▃·▁ beats 8/17 very hard · avg 11% 2.1% at its level
MathArena BrokenArXiv April 2026 … 23.8% ▂▂▆●▃▃▂▂▂··▂ beats 9/18 very hard · avg 0.8% 1.6% sharp · noisy
AIME 2025 88.9% ▁·▁▁▁▂▅▆█●▅▁ beats 109/133 medium · avg 48% 1.2% sharp · maxed out · small run
HMMT 2025 (edition unspecified) 95.2% ▃█▇▅▇▅▁·●··▁ beats 31/32 hard · avg 31% 1.1% sharp · small run
MathArena BrokenArXiv May 2026 (5… 19.5% ██·▄●▄····▄▄ beats 5/8 very hard · avg 1.9% 1.1% sharp
MathArena BrokenArXiv March 2026 … 19.6% █·▆█●·▂▂···▂ beats 11/14 very hard · avg 1.0% 0.9% sharp · noisy
MathArena BrokenArXiv February 20… 17.7% ▃▂▂▃█●▂·▃··▂ beats 11/16 very hard · avg 0.9% 0.9% sharp · noisy
HMMT February 2026 94.7% █▄▄········● beats 5/5 hard · avg 31% 0.7% sharp · noisy
web search · #74 of 84 · 2.7% of its score
Trails [GPT-4.1 nano](gpt-4-1-nano-2025-04-14.md), [GPT-4o mini (2024-07-18)](gpt-4o-mini-2024-07-18.md), [Mistral Small 3.1 24B (2503)](mistral-small-3-1.md) here, though they rank lower overall.
test result vs the others in its run difficulty counts why
Humanity's Last Exam with tools 45.1% ▄····●·█·▄·▄ beats 1/5 hard · avg 15% 1.5% at its level
BrowseComp 46.9% ▄·▄●···▄··█▄ beats 2/6 hard · avg 18% 1.2% at its level
other tests (15)
test result vs the others in its run difficulty counts why
GPQA Diamond 88.9% █···██·█··█● beats 5/5 hard · avg 41% 6.0% sharp · several runs
Humanity's Last Exam 32.2% ▄··█▄··●·▄·▄ beats 4/6 very hard · avg 5.7% 5.4% at its level · several runs
GPQA (subset unspecified) 88.5% ▄█▄▄·●·····▄ beats 5/6 hard · avg 47% 4.1% sharp
MMLU-Pro 86.4% ▄·····▄●▄·▄█ beats 2/6 medium · avg 53% 3.9% several runs
DSEffi-Bench Pass (1,000 data-sci… 57.7% ▅··▃▅·█·▃●█▃ beats 11/15 hard · avg 23% 2.1% at its level
MultiChallenge 63.5% ▁··········● beats 3/6 hard · avg 28% 2.0% at its level · several runs
EQ-Bench 3 (emotional intelligenc… 1287.2 ▁·▂▁▄▅▅▃●▄▁▁ beats 42/62 avg 1034 1.8% typical
IFBench (aggregation unspecified) 80.3% ▄···▄▄··▄··● beats 5/5 medium · avg 43% 1.7% several runs
LongBench v2 57.0% ▂·▁·▂▃▁▄▄▃●▄ beats 21/33 very hard · avg 34% 1.5% at its level
MMLU-ProX 84.3% ▄·····▄·●▄█▄ beats 2/6 medium · avg 58% 1.3% typical
IFBench prompt-level (loose) 82.0% ▄▄·▄··▄▄···● beats 6/6 medium · avg 47% 1.1% typical
RobloxQA v2.0 multiple choice (To… 90.6% ▁▁▃▁●▄▃▃▆▂▂▂ beats 16/49 easy · avg 77% 0.9% typical
SkillsBench 87-task subset 53.5% ▄··▄▄▄···█·● beats 6/7 very hard · avg 9.1% 0.9% at its level
WideSearch corrected dataset 74.4% ▃······▃▃▃●▃ beats 5/7 hard · avg 22% 0.8% at its level · noisy
CritPt 10.6% ▄····▄█···●▄ beats 5/6 very hard · avg 1.1% 0.6% too hard for it · noisy
source results share of the score
NVIDIA 18 32% ▇▇▇▇
MathArena ArXivMath / BrokenArXiv / ArX… 8 19% ▇▇
Solar Open 2 Technical Report 13 18% ▇▇
ai-sage/GigaChat3.5-432B-A28B-Reasoning… 10 12% ▇
LMArena leaderboard dataset 2 8% ▇
Roblox LLM Leaderboard 2 3% ▇
DSEffi-Bench: Demystifying Large Langua… 1 2% ▇
EQ-Bench 3 canonical leaderboard results 1 2% ▇
GameLogicBench: Evaluating Coding Agent… 1 2% ▇
inclusionAI/Ling-3.0-flash model card 2 2% ▇
prior: pull toward the middle 0% ▇
weighted average = 70.4
Licences: NVIDIA ([CC BY 4.0](../../licenses/report-2606.15007v1.txt)) · MathArena ArXivMath / BrokenArXiv / ArXivLean model-output datasets ([CC BY-SA 4.0](../../licenses/matharena-sharealike-NOTICE.txt)) · Solar Open 2 Technical Report ([CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/)) · ai-sage/GigaChat3.5-432B-A28B-Reasoning model card ([MIT](https://huggingface.co/ai-sage/GigaChat3.5-432B-A28B-Reasoning/blob/02f55379ac0b20abd9378791bffd6254725e6f5c/README.md)) · LMArena leaderboard dataset ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · Roblox LLM Leaderboard ([MIT](../../licenses/roblox-llm-leaderboard-MIT.txt)) · DSEffi-Bench: Demystifying Large Language Models' Capability in Efficient Data Science Code Generation ([CC BY 4.0](../../licenses/paper-2608.30248v1-CC-BY-4.0.txt)) · EQ-Bench 3 canonical leaderboard results ([MIT](../../licenses/eqbench3-MIT.txt)) · GameLogicBench: Evaluating Coding Agents on Runtime Game Logic with Tick-Level State Assertions ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · inclusionAI/Ling-3.0-flash model card ([MIT](../../licenses/cards-final-NOTICE.txt))
## 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 V4 Flash (Preview, Apr 2026) (ud-q4_k_xl, needs ~155.1 GB, from [huggingface.co/unsloth](https://huggingface.co/unsloth/DeepSeek-V4-Flash-GGUF/tree/e3aa0d6a5fa4f820d9e132ac1fd1d01e1b2b49e0))
```sh
llama cli -hf unsloth/DeepSeek-V4-Flash-GGUF:UD-Q4_K_XL
```
3. Or start a local server and open the address it prints (http://localhost:9931)
```sh
llama serve -hf unsloth/DeepSeek-V4-Flash-GGUF:UD-Q4_K_XL
```
Other versions: ud-q2_k_xl (~96.83 GB) `llama cli -hf unsloth/DeepSeek-V4-Flash-GGUF:UD-Q2_K_XL`
### 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 V4 Flash (Preview, Apr 2026) (ud-q4_k_xl, needs ~155.1 GB, from [huggingface.co/unsloth](https://huggingface.co/unsloth/DeepSeek-V4-Flash-GGUF/tree/e3aa0d6a5fa4f820d9e132ac1fd1d01e1b2b49e0))
```sh
lms get https://huggingface.co/unsloth/DeepSeek-V4-Flash-GGUF@UD-Q4_K_XL -y
```
3. Then run lms ls to see its name, and lms chat followed by that name
Other versions: ud-q2_k_xl (~96.83 GB) `lms get https://huggingface.co/unsloth/DeepSeek-V4-Flash-GGUF@UD-Q2_K_XL -y`
Model repository and documentation: [huggingface.co/deepseek-ai](https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash)
[interactive version](https://whatmodel.sh/model/deepseek-v4-flash-preview) · [JSON](https://whatmodel.sh/api?format=json&all=1)