DeepSeek V3.1 Thinking
Thinking enabled version of Deepseek V3.1. DeepSeek-V3.1 is a hybrid model that supports both thinking mode and non-thinking mode. It does better at tool calling and agent tasks, and has higher thinking efficiency than its predecessor. Quantized at FP8.
Pricing
Auto routing · per 1M tokens- Input
- $0.20
- Output
- $0.70
- Cache read
- $0.10
Specifications
- Context window
- 128K
- Max output
- 65.5K
- Parameters
- 671B / 37B
- Total / active
Benchmarks
Benchmarks
Sourced from LMArena.
Arena Score
1415.9
Overall Rank
#142 / 421
Votes
11,441
Confidence Interval
1409.2 - 1422.5
Category Scores
Coding
#155 / 416
1,873 votes
1455.1
Math
#139 / 404
658 votes
1414.5
Longer Query
#100 / 399
2,375 votes
1446.1
Creative Writing
#108 / 419
1,618 votes
1401.5
Instruction Following
#120 / 421
2,829 votes
1416.3
Hard Prompts
#141 / 421
5,079 votes
1436.7
Additional Categories22
Industry Mathematical
#106 / 401
562 votes
1442.6
Chinese
#112 / 395
1,141 votes
1469.8
Industry Writing And Literature And Language
#113 / 420
2,720 votes
1407.7
Industry Medicine And Healthcare
#115 / 390
685 votes
1449.2
Industry Entertainment And Sports And Media
#125 / 419
2,198 votes
1393.2
Japanese
#126 / 288
258 votes
1368.7
Spanish
#134 / 303
386 votes
1410.4
German
#136 / 320
322 votes
1400.4
Industry Business And Management And Financial Operations
#136 / 414
1,937 votes
1417.7
Hard Prompts English
#137 / 419
2,127 votes
1446.4
English
#138 / 421
4,654 votes
1430.0
Polish
#142 / 244
786 votes
1399.1
Non English
#143 / 421
6,787 votes
1400.6
Exclude Ties
#144 / 421
8,132 votes
1402.8
Industry Legal And Government
#146 / 393
770 votes
1422.2
Expert
#147 / 371
536 votes
1433.4
Multi Turn
#147 / 419
1,815 votes
1413.9
Industry Software And It Services
#149 / 421
3,457 votes
1447.4
Korean
#152 / 289
374 votes
1345.7
Russian
#153 / 385
725 votes
1401.8
Industry Life And Physical And Social Science
#155 / 419
1,829 votes
1427.9
French
#165 / 301
183 votes
1406.3
Published 2026-11-09 · Matched as DeepSeek-V3.1 (Thinking) · CC BY 4.0
LMArena DatasetLeaderboard results
These leaderboards tested DeepSeek V3.1, 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 | 140.8 | #93 of 115 |
Creative Writing v3 | 1436 | #63 of 96 |
Longform Writing | 62.3 | #53 of 97 |
UGI Leaderboard UGI Leaderboard data, Apache-2.0 · Updated Oct 16, 2026 | ||
UGI Score | 51.4 | #35 of 149 |
UGI Willingness | 4.5 | #52 of 149 |
UGI Writing | 51.9 | #58 of 149 |
Filtered and ranked to models on NanoGPT.
Providers
Auto routing is available for this model. Explicit provider selection is not available.
 Recommended | $0.20 | $0.70 | $0.10 | 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.