DeepSeek-V3.2-Exp
DeepSeek-V3.2-Exp is an open-weights AI model from DeepSeek with an index score of 140 (GPT-4 = 100), rank 97 of 217.
140124–157Index score#97 of 217 models
$0.27/ $0.41Price per million tokensinput / output · via OpenRouter
164KContext windowtokens it reads at once
29 Sep 2025Released1 year 2 months ago
Where it sitson the frontier
Gemini 2.5 Pro reached this score 6 months earlier.
Test resultsby area · tick = best by any model
Reasoning & knowledgeevidence
Mathsno evidence yet
Codingevidence
Agentsevidence
Professional workno evidence yet
Novel problemsno evidence yet
Long tasksno evidence yet
Third-party results only, at the model’s best standard setting. Hover a bar for who ran it.
What others sayrank on 9 leaderboards
The Model Indexthird-party tests, our method#97of 217 · 140
LMArena · Creative writingcreative writing#80of 360 · 1408LMArena · Instruction followingfollowing detailed instructions#100of 362 · 1416LMArena · Long promptslong, complex requests#94of 340 · 1441LMArena · Textoverall chat quality, judged by people#102of 362 · 1425LMArena · Codingcoding questions in chat#102of 357 · 1475LMArena · Mathsmaths questions in chat#106of 346 · 1421Epoch Capabilities Indexoverall capability from many benchmarks#97of 231 · 144.4LMArena · WebDevbuilding web apps and front ends#116of 133 · 1272Best attop three among current models
Not in the top three among current models on any category we track.
Where to use itprices checked 18 Nov 2026
API providers 1 · per 1M tokens, in / out
Run it yourself
Official weightsdeepseek-ai/DeepSeek-V3.2-Exp↗Host prices from OpenRouter; going direct can differ.
Facts and sources
Released29 Sep 2025
Announcementapi-docs.deepseek.com ↗
Release entryDeepSeek-V3.2-Exp (sparse attention)
Scored asDeepSeek-V3.2-Exp
Output speed–
Hugging Facedeepseek-ai/DeepSeek-V3.2-Exp ↗