DeepSeek-V3.2
DeepSeek-V3.2 (DeepSeek) was released on Dec 1, 2025. In the Epoch AI capability ranking (ECI), it ranks #83 out of 267 models, with 146.11. On Arena, which measures human preference in blind comparisons, it has 1,455 ELO points (#120). Against models of the same generation, its best result is on AIME 2024 and 2025 (OTIS): 87.8%, #12 of 58.
Where DeepSeek-V3.2 stands out and where it falls behind, by the numbers
DeepSeek-V3.2 has published results on 15 tests compiled by Epoch AI. The yardstick here is its generation: its position among the models released between Mar 6, 2025 and Mar 1, 2026 that took the same test. Against the whole archive, which includes 2023 models, a recent release lands near the top of almost everything, and that says nothing.
Where it stands out
Tests where it sits in the top third of its own generation.
Where it falls behind
Tests where it sits in the bottom half of its own generation.
See all 15 tests, one by one
| Test | Score | In its generation | Overall | Test record |
|---|---|---|---|---|
| GPQA Diamond | 83.4% | #18 / 57 | #72 / 173 | 96.4% Gemini 4 Argon |
| AIME 2024 and 2025 (OTIS) | 87.8% | #12 / 58 | #64 / 168 | 100.0% GPT-5.5 |
| Dtbench | 85.6% | #13 / 54 | #55 / 148 | 98.9% Claude Opus 5.5 |
| Lmca | 28.82 | #25 / 49 | #79 / 121 | 68.23 Claude Opus 5.5 |
| ARC-AGI 2 | 4.0% | #26 / 39 (tied with 1) | #62 / 83 | 95.0% GPT-6 Astra |
| WebDev Arena | 1,325 | #22 / 25 | #76 / 82 | 1,820 Claude Opus 5.5 |
| ARC-AGI | 57.0% | #20 / 39 | #54 / 81 | 98.5% Claude Fable 5 |
| ProofBench | 8.0% | #13 / 15 | #66 / 74 | 100.0% Claude Fable 5.1 |
| FrontierMath | 22.1% | #12 / 38 | #26 / 65 | 51.7% GPT-5.5 |
| Apex Agents | 21.3% | #5 / 6 | #49 / 50 | 82.4% Gemini 4 Argon |
| FrontierMath tier 4 | 2.1% | #19 / 32 (tied with 1) | #32 / 48 | 39.6% GPT-5.5 Pro |
| Aider Polyglot | 74.20 | #6 / 25 (tied with 1) | #6 / 38 | 88.00 GPT-5 |
| Terminal-Bench | 39.6% | #16 / 28 | #20 / 33 | 84.7% GPT-5.5 |
| Cl Bench | 12.4% | #16 / 17 | #18 / 19 | 27.9% GPT-5.4 |
| Cl Bench Life | 7.4% | #9 / 10 | #12 / 13 | 22.2% GPT-5.5 |
For each test, the model's best effort level counts, as in the ranking. Source of the scores: Epoch AI, under CC BY 4.0.
What to compare DeepSeek-V3.2 with
The models from other labs closest in capability and released in the same period, at most two per lab. Click any of them to see its page.
| Model | Lab | ECI | ELO | LiveBench | Release |
|---|---|---|---|---|---|
| Grok 4 | xAI | 146.43 | n/a | n/a | Jul 9, 2025 |
| Qwen 3.5 Plus (hosted 397B-A17B) | Alibaba | 146.32 | n/a | n/a | Feb 16, 2026 |
| Qwen3.5-122B-A10B | Alibaba | 146.26 | 1,466 | n/a | Feb 24, 2026 |
| GPT-5.4 Nano | OpenAI | 146.12 | 1,404 | 69.6 | Mar 17, 2026 |
| DeepSeek-V3.2 | DeepSeek | 146.11 | 1,455 | n/a | Dec 1, 2025 |
| Kimi K2 Thinking | Moonshot | 145.96 | n/a | n/a | Nov 6, 2025 |
| MiniMax-M2.7 | MiniMax | 145.88 | 1,454 | n/a | Mar 18, 2026 |
DeepSeek-V3.2 or GPT-5.4 Nano?
GPT-5.4 Nano is the model from another lab closest to DeepSeek-V3.2 in capability. Across the 9 tests both took, this is the score:
| Test | DeepSeek-V3.2 | GPT-5.4 Nano |
|---|---|---|
| FrontierMath tier 4 | 2.1% | 6.3% |
| ProofBench | 8.0% | 5.0% |
| ARC-AGI 2 | 4.0% | 5.7% |
| Lmca | 28.82 | 36.86 |
| FrontierMath | 22.1% | 25.9% |
| ARC-AGI | 57.0% | 51.5% |
| Dtbench | 85.6% | 80.3% |
| GPQA Diamond | 83.4% | 78.5% |
| AIME 2024 and 2025 (OTIS) | 87.8% | 87.8% |
Unweighted score: each test counts one point, whether math or coding. The dot marks who did better on each row.
Other models from DeepSeek
The models from the same lab closest in time, newest first.
| Model | Release | ECI | ELO | LiveBench | Access |
|---|---|---|---|---|---|
| DeepSeek V4 Flash 0731 | Jul 31, 2026 | 154.49 | n/a | 74.2 | Open weights (unrestricted) |
| DeepSeek-V4-Pro | Apr 24, 2026 | 149.19 | 1,486 | 71.6 | Open weights (unrestricted) |
| DeepSeek v4 (unknown) | Apr 24, 2026 | 146.26 | n/a | n/a | Open weights (unrestricted) |
| DeepSeek-V4-Flash | Apr 24, 2026 | 146.05 | 1,468 | 65.5 | Open weights (unrestricted) |
| DeepSeek-V3.2 | Dec 1, 2025 | 146.11 | 1,455 | n/a | Open weights (unrestricted) |
| DeepSeek-V3.2-Exp | Sep 29, 2025 | 144.97 | 1,461 | n/a | Open weights (unrestricted) |
| DeepSeek-V3.1 | Aug 21, 2025 | 139.91 | 1,469 | n/a | Open weights (unrestricted) |
| DeepSeek-R1 (May 2025) | May 28, 2025 | 141.27 | 1,400 | n/a | Open weights (unrestricted) |
| DeepSeek-V3 (Mar 2025) | Mar 24, 2025 | 135.93 | 1,346 | n/a | Open weights (restricted use) |
Have you used DeepSeek-V3.2? Rate it
Users' opinion, not measurement: the rating is kept apart from the ranking and does not move anyone's position. One vote per browser, no sign-up, and you can change your rating later.
What people ask about DeepSeek-V3.2
Where does DeepSeek-V3.2 rank among AI models?
In the ROO3 AI Benchmark, ordered by the Epoch AI ECI, DeepSeek-V3.2 ranks #83 out of 267 models, with an ECI of 146.11. The ECI is a composite index that combines several hard benchmarks into a single scale. The position changes when a new model comes out, and the page is refreshed six times a day.
Is DeepSeek-V3.2 better than GPT-5.4 Nano?
Across the 9 tests both models took and Epoch AI compiled, DeepSeek-V3.2 comes out ahead on 5, behind on 4 and tied on 0. It is a test-by-test count with no weighting: a win in math counts the same as a win in coding. To choose between the two, look at the tests that resemble your own use.
What is DeepSeek-V3.2 best at?
Against models of the same generation (released up to 9 months before and up to 3 months after it), the best relative results of DeepSeek-V3.2 are: AIME 2024 and 2025 (OTIS) (#12 of 58); Dtbench (#13 of 54); Aider Polyglot (#6 of 25). The yardstick is its position among the models of that period that took the same test.
Where do the numbers on this page come from?
From the same three sources as the ranking, all openly licensed: Epoch AI (ECI and individual tests, CC BY 4.0), Arena (human-preference ELO, CC BY 4.0) and LiveBench (overall average, CC BY-SA 4.0). ROO3 does not measure any model. When the page shows a figure published by the maker itself, it appears in the written analysis, labeled as the maker's, and never enters the ranking.
Where each number comes from
Epoch AI, "AI Benchmarking Hub". Published at epoch.ai. Data under the CC BY 4.0 licence, reordered by roo3.co.
The Arena (LMArena) leaderboard, from the public lmarena-ai/leaderboard-dataset on Hugging Face, under the CC BY 4.0 licence. We are not affiliated with the Arena.
LiveBench, by the LiveBench team, published at livebench.ai under the CC BY-SA 4.0 license. The overall average is recomputed by roo3.co with the same formula as the website (average of the categories), and those numbers stay under CC BY-SA 4.0. We are not affiliated with LiveBench.
← See the full ranking