Compare / head-to-head
DeepSeek V3.2-Specialevs
Gemini 3 Pro Preview
Gemini 3 Pro Preview leads 3 of 4 shared benchmarks.
Benchmarks from cited public sources; pricing from official pages; status from official provider feeds.
Shared benchmarks
1 – 3
Gemini 3 Pro Preview leads
Cheaper per token
—
list price, input + output
Larger context
—
unknown
Providers
2 providers
DeepSeek · Google
DeepSeek V3.2-Speciale
DeepSeek · released 2025-12-01
text
- Context
- —
- Max out
- —
- Input /1M
- —
- Output /1M
- —
- Cached /1M
- —
- Scores
- 9 · 4 core
Gemini 3 Pro Preview
Google · released 2025-11-18
textvisionaudiovideo
Quality
Benchmark matrix
4 shared · 5 only DeepSeek V3.2-Speciale · 20 only Gemini 3 Pro Preview| Benchmark | DeepSeek V3.2-Speciale | Gemini 3 Pro Preview | Δ | Edge |
|---|---|---|---|---|
| Reported by both · 4 | ||||
| GPQA Diamond | 85.7% ↗ | 91.9% ↗ | -6.2 pt | Gemini 3 Pro Preview |
| AIME 2025 | 96% ↗ | 95% ↗ | +1 pt | DeepSeek V3.2-Speciale |
| SimpleBench (SimpleBench) | 52.6% ↗ | 76.4% ↗ | -23.8 pt | Gemini 3 Pro Preview |
| WeirdML (Håvard Tveit Ihle) | 46.7% ↗ | 69.9% ↗ | -23.2 pt | Gemini 3 Pro Preview |
| Only DeepSeek V3.2-Speciale reports · 5 | ||||
| HLE (text-only) | 30.6% ↗ | not reported | — | — |
| HMMT Feb 2025 | 99.2% ↗ | not reported | — | — |
| HMMT Nov 2025 | 94.4% ↗ | not reported | — | — |
| IMOAnswerBench | 84.5% ↗ | not reported | — | — |
| LiveCodeBench (Pass@1-COT) | 88.7% ↗ | not reported | — | — |
| Only Gemini 3 Pro Preview reports · 20 | ||||
| SWE-bench Verified | not reported | 76.2% ↗ | — | — |
| Humanity's Last Exam (no tools) | not reported | 37.5% ↗ | — | — |
| AIME 2025 (with code execution) | not reported | 100% ↗ | — | — |
| ARC-AGI-2 (ARC Prize Verified) | not reported | 31.1% ↗ | — | — |
| BALROG (BALROG) | not reported | 58.1% ↗ | — | — |
| CharXiv Reasoning | not reported | 81.4% ↗ | — | — |
| Chess Puzzles (Epoch AI run) | not reported | 31% ↗epoch run | — | — |
| GSO Opt@1 (GSO) | not reported | 17.7% ↗ | — | — |
| Humanity's Last Exam (with search and code execution) | not reported | 45.8% ↗ | — | — |
| MathArena Apex | not reported | 23.4% ↗ | — | — |
| MMMLU | not reported | 91.8% ↗ | — | — |
| MMMU-Pro | not reported | 81% ↗ | — | — |
| OTIS Mock AIME 2024-2025 (Epoch AI run) | not reported | 91.4% ↗epoch run | — | — |
| SAGE (Vals AI) | not reported | 47.6% ↗ | — | — |
| SimpleQA Verified | not reported | 72.1% ↗ | — | — |
| SWE-bench Verified (Epoch AI run) | not reported | 72.9% ↗epoch run | — | — |
| Terminal-Bench 2.0 (Terminus-2 agent) | not reported | 54.2% ↗ | — | — |
| Vending-Bench 2 (Andon Labs) | not reported | 5478.16 ↗ | — | — |
| Video-MMMU | not reported | 87.6% ↗ | — | — |
| τ2-bench | not reported | 85.4% ↗ | — | — |
Scores tagged "epoch" or "matharena" are independent runs, used only where the lab has not published its own; ⚠ marks rows MathArena flags as released after the competition. Higher is better on every row. Δ is DeepSeek V3.2-Speciale minus Gemini 3 Pro Preview in the benchmark's own unit. "Not reported" means the lab has not published that figure; it is not a zero. ↗ opens the source.
Specs & pricing
Side by side
Official pricing pages and model cards| Spec | DeepSeek V3.2-Speciale | Gemini 3 Pro Preview | Edge |
|---|---|---|---|
| Context window | — | 1.0M tokens | — |
| Max output | — | 66K tokens | — |
| Input price / 1M | — | $2 ↗ | — |
| Output price / 1M | — | $12 ↗ | — |
| Cached input / 1M | — | — | — |
| Input + output / 1M Lower is cheaper. List prices; batch, tool and regional fees excluded. | — | $14 | — |
| Modalities | text | text · vision · audio · video | Gemini 3 Pro Preview |
| Released | 2025-12-01 | 2025-11-18 | — |
| Cited benchmark scores | 9 | 25 | — |
Reliability
Provider status
Live from /statusMore matchups
DeepSeek V3.2-Speciale vs …
Models sharing the most benchmarksMore matchups
Gemini 3 Pro Preview vs …
Models sharing the most benchmarksBuilt by Respan
Which one wins on your data?
Public benchmarks are a starting point. Run DeepSeek V3.2-Speciale and Gemini 3 Pro Preview on your own prompts with Respan evals, or route to either through one gateway key with automatic failover.