whatmodel.sh / DeepSeek V3.2 Exp

DeepSeek V3.2 Exp

+ contribute

DeepSeek V3.2 Exp is an open-weights language model from DeepSeek, released September 2025. Its published size is 671B parameters (37B active per token). Licence: mit.

Catalogue identity notes

Comparison-only. Arena entries 'deepseek-v3.2-exp-thinking' (text), 'deepseek-v3.2-exp' (text, webdev). Maker repo deepseek-ai/DeepSeek-V3.2-Exp @194c67e: licence tag mit. Parameters: same DeepSeek-V3 structure as V3.1-Terminus, 671B / 37B (HF safetensors metadata 685.4B incl. MTP). Released 2025-09: month the maker's Hugging Face repo was created (createdAt 2025-09-29). Experimental release, distinct from the catalogued deepseek-v3-2.

66.5±1.0#122 of 719 LLMsfrom 14 benchmark results across 6 sources

$ how it compares

The LLMs ranked either side of DeepSeek V3.2 Exp, out of 828 with a score, then more from DeepSeek.

#	model	score	where	RAM	released
139	Qwen3.5 35B-A3B Alibaba	67±0	api $0.35/M +local	~23GB	2026-02
140	DeepSeek V3.1 Terminus DeepSeek	67±1	local	~230GB	2025-09
141	DeepSeek V3.2 Exp DeepSeek	67±1	local	~726GB	2025-09
142	Qwen3 235B-A22B Thinking 2507 Alibaba	66±1	api $0.97/M +local	~76GB	2025-07
143	Grok 4.1 Fast (Reasoning) SpaceXAI (xAI)	66±1	api $0.28/M		2025-11
	more from DeepSeek
21	DeepSeek V4.1 Flash DeepSeek	78±1	api $0.26/M +local	~537GB	2026-09
31	DeepSeek V4 Pro (0813) DeepSeek	76±1	api $1.63/M +local	~1010GB	2026-08
35	DeepSeek V3.2 Speciale DeepSeek	76±3	comparison only

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▆▅▅▅▄▃▁ #139 of 300 * · 4.1% of its score · 1 test

Trails SmolLM3 3B, Jamba 1.5 Large, GPT-4o (2024-05-13) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
BFCL V4 (December 2025 archive)	54.1% ✎	▁▁▁▂██▅▅▆▂▂▂ beats 63/77	hard · avg 32%	▇▇ 4.1%	at its level

coding▁▁▁▃▆██▇▆▄▃▁ #134 of 458 · 43% of its score · 4 tests

Beats Qwen3.5 Plus, Gemini 2.5 Pro, Gemini 3.1 Flash-Lite here, though they rank higher overall. Trails Devstral Small 2 24B, Llama 3.1 Nemotron Ultra 253B v1, Nemotron 3.5 Lightning 30B-A3B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
Aider polyglot (225 exercises, two attempts)	74.2% ✎	▂·▂▃▅▂▃█▆▄▂▂ beats 43/49	very hard · avg 8.5%	▇▇▇▇▇▇▇▇▇ 22%	sharp · at its level
Arena WebDev (Code Arena)	1271.8 ✎	▁▂▄▃▅█▄▇▆▄▁▁ beats 17/119	avg 981	▇▇▇▇▇▇ 15%	sharp
WebApp1K-Duo-React pass@1 (onekq-ai)	64.1% ✎	▁··▁·▁▁▁▆█▇█ beats 19/45	hard · avg 32%	▇▇ 3.8%	at its level
LiveCodeBench v6 (Aug 2024–May 2025)	71.6% ✎	▄▄·▄·····█·▄ beats 5/5	medium · avg 45%	▇ 3.1%	at its level

chat & writing▁▁▁▃▅▄▅█▆▆▇▃ #83 of 393 * · 8.3% of its score · 1 test

Beats LongCat 2.0 (1.6T-A48B), Inkling Small (266B-A12B), Qwen3.5 122B-A10B here, though they rank higher overall. Trails Gemma 4 12B, Gemma 4 26B-A4B, Mistral Medium 3.5 (128B) here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
LMArena (text)	1424.9 ✎	▁▁▁▂▃▃▃▆▆▇█▃ beats 276/379	avg 1268	▇▇▇ 8.3%	blunt

math▁▂▃▅▆▅▄█▄▁▁▁ #72 of 398 · 15% of its score · 3 tests

Beats Claude Sonnet 4.6, Claude Opus 4.5, Qwen3.7 Plus here, though they rank higher overall. Trails Granite 4.2 30B, INTELLECT-3, EXAONE 4.5 33B here, though they rank lower overall.

test	result	vs the others in its run	difficulty	counts	why
AIME 2024	88.1% ✎	█···███·█··█ beats 4/5	medium · avg 49%	▇▇▇ 8.0%	sharp
AIME 2025	84.7% ✎	█·█···███··█ beats 4/5	medium · avg 43%	▇▇▇ 6.7%	sharp
MATH-500	96.8% ✎	▄······▄▄▄·█ beats 2/5	very easy · avg 86%	▇ 0.8%	maxed out · noisy

other tests5 not in a task · 29% of its score

test	result	vs the others in its run	difficulty	counts	why
GPQA Diamond	81.4% ✎	█··██·██···█ beats 5/5	hard · avg 42%	▇▇▇▇▇▇ 15%	sharp · at its level
Humanity's Last Exam	17.9% ✎	▄····▄·█·▄·▄ beats 5/5	very hard · avg 4.3%	▇▇▇▇ 9.1%	typical
MMLU-Pro	84.6% ✎	▁▁·▁▁▁▃▂▄▅█▁ beats 166/188	medium · avg 51%	▇▇ 4.3%	typical
Vectara hallucination leaderboard, 2025 dataset (HHEM-2.3, factual consistency rate)	95.4% ✎	▁▁▂▃▄▄█▆▃▂▂▁ beats 73/140	very easy · avg 94%	▇ 0.2%	blunt · maxed out · noisy
Vectara hallucination leaderboard (HHEM-2.3, factual consistency rate)	94.7% ✎	▁▂▁▂▅█▂▃▄▃▁▁ beats 86/98	very easy · avg 90%	▇ 0.0%	blunt

$ where to run it

Other ways to run it

version

Run it as published · needs ~689.48 GB, NVIDIA only · from huggingface.co/deepseek-ai
python -m sglang.launch_server --model deepseek-ai/DeepSeek-V3.2-Exp --tp 8 --dp 8 --enable-dp-attention

Weights and documentation: huggingface.co/deepseek-ai ↗

$ how the score was worked out · Prime Intellect INTELLECT-3 m… 47%, LMArena leaderboard dataset 23%, Aider polyglot benchmark runs 22% and 3 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	licence
Prime Intellect INTELLECT-3 model card: Evaluation table (Environments Hub runs)	7	▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 47%	MIT
LMArena leaderboard dataset	2	▇▇▇▇▇▇▇ 23%	CC BY 4.0
Aider polyglot benchmark runs	1	▇▇▇▇▇▇ 22%	Apache-2.0
Berkeley Function Calling Leaderboard result archive	1	▇ 4%	Apache-2.0
WebApp1K models leaderboard (onekq-ai)	1	▇ 4%	MIT
Vectara hallucination leaderboard (vectara/hallucination-leaderboard repository)	2	▇ 0%	Apache-2.0
prior · pull toward the middle		▇ 0%
weighted average		= 66.5

$ improve this page

Update model information Add a place to run it Add a test result Propose a benchmark Correct a benchmark score Edit data on GitHub

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 2026-10-11 02:20 UTC · sources and licences · json