DeepSeek V3.2 Exp Thinking
DeepSeek V3.2 Exp Thinking is an experimental V3-series reasoning model with improved long-context efficiency compared with earlier V3 releases. FP8.
- Reasoning
Pricing
Auto routing · per 1M tokens- Input
- $0.28
- Output
- $0.42
- Cache read
- $0.14
Specifications
- Context window
- 163.8K
- Max output
- 65.5K
- Parameters
- 685B / 37B
- Total / active
Benchmarks
Benchmarks
Sourced from LMArena.
Arena Score
1425.0
Overall Rank
#124 / 421
Votes
8,963
Confidence Interval
1418.5 - 1431.6
Category Scores
Coding
#122 / 416
1,923 votes
1474.9
Math
#129 / 404
474 votes
1421.1
Longer Query
#123 / 399
2,098 votes
1433.7
Creative Writing
#129 / 419
1,186 votes
1390.1
Instruction Following
#122 / 421
2,550 votes
1415.8
Hard Prompts
#124 / 421
4,699 votes
1446.2
Additional Categories21
Industry Medicine And Healthcare
#70 / 391
442 votes
1474.8
German
#100 / 320
184 votes
1436.3
Polish
#103 / 243
397 votes
1429.6
Hard Prompts English
#112 / 419
2,053 votes
1459.8
Industry Life And Physical And Social Science
#114 / 419
1,415 votes
1450.9
Industry Mathematical
#116 / 402
399 votes
1436.9
Chinese
#118 / 395
515 votes
1465.2
English
#119 / 421
3,874 votes
1439.8
Expert
#121 / 371
418 votes
1451.6
Russian
#125 / 385
648 votes
1413.9
Exclude Ties
#127 / 421
6,394 votes
1415.4
Industry Entertainment And Sports And Media
#128 / 419
1,656 votes
1389.6
Industry Software And It Services
#128 / 421
3,190 votes
1459.3
Non English
#130 / 421
5,089 votes
1406.3
Industry Writing And Literature And Language
#136 / 420
2,089 votes
1398.0
Multi Turn
#139 / 419
1,396 votes
1419.5
Industry Legal And Government
#140 / 393
499 votes
1425.2
Spanish
#140 / 304
311 votes
1407.7
Korean
#142 / 289
215 votes
1351.5
French
#151 / 301
147 votes
1418.7
Industry Business And Management And Financial Operations
#155 / 414
1,778 votes
1407.8
Published 2026-11-11 · Matched as DeepSeek-V3.2-Exp (Thinking) · CC BY 4.0
LMArena DatasetLeaderboard results
These leaderboards tested DeepSeek V3.2, another version of this model.
| Benchmark | Score | Rank on NanoGPT |
|---|---|---|
Epoch AI Benchmarking Hub Epoch AI data, CC BY 4.0 · Updated Nov 13, 2026 | ||
Epoch Capabilities Index | 146.3 | #79 of 115 |
Creative Writing v3 | 1515 | #55 of 96 |
Longform Writing | 66.8 | #46 of 97 |
UGI Leaderboard UGI Leaderboard data, Apache-2.0 · Updated Oct 16, 2026 | ||
UGI Score | 57.8 | #13 of 149 |
UGI Willingness | 7.2 | #9 of 149 |
UGI Writing | 57.3 | #43 of 149 |
Filtered and ranked to models on NanoGPT.
Providers
Auto routing is available for this model. Explicit provider selection is not available.
 Recommended | $0.28 | $0.42 | $0.14 | N/A | N/A |
Batch rows for this model use normal routing and cost the same as standard API requests. Set provider in a row to choose a provider.