Compare / head-to-head
DeepSeek V3.1vs
Qwen3.5 2B
DeepSeek V3.1 leads 2 of 2 shared benchmarks. Qwen3.5 2B has the larger context window (262K tokens).
Benchmarks from cited public sources; pricing from official pages; status from official provider feeds.
Shared benchmarks
2 – 0
DeepSeek V3.1 leads
Cheaper per token
—
list price, input + output
Larger context
Qwen3.5 2B
262K tokens
Providers
2 providers
DeepSeek · Qwen
DeepSeek V3.1
DeepSeek · released 2025-08-21
text
- Context
- 128K
- Max out
- —
- Input /1M
- —
- Output /1M
- —
- Cached /1M
- —
- Scores
- 18 · 9 core
Qwen3.5 2B
Qwen · released 2026-03-02
textvisionvideo
- Context
- 262K
- Max out
- —
- Input /1M
- —
- Output /1M
- —
- Cached /1M
- —
- Scores
- 16 · 5 core
Quality
Benchmark matrix
2 shared · 16 only DeepSeek V3.1 · 14 only Qwen3.5 2B| Benchmark | DeepSeek V3.1 | Qwen3.5 2B | Δ | Edge |
|---|---|---|---|---|
| Reported by both · 2 | ||||
| MMLU-Pro (thinking) | 84.8% ↗ | 66.5% ↗ | +18.3 pt | DeepSeek V3.1 |
| MMLU-Redux (thinking) | 93.7% ↗ | 79.6% ↗ | +14.1 pt | DeepSeek V3.1 |
| Only DeepSeek V3.1 reports · 16 | ||||
| LMArena Elo | 1417.1 ↗ | not reported | — | — |
| Aider-Polyglot (thinking) | 76.3% ↗ | not reported | — | — |
| AIME 2024 (thinking) | 93.1% ↗ | not reported | — | — |
| AIME 2025 (thinking) | 88.4% ↗ | not reported | — | — |
| BrowseComp | 30% ↗ | not reported | — | — |
| GPQA-Diamond (thinking) | 80.1% ↗ | not reported | — | — |
| HMMT 2025 (thinking) | 84.2% ↗ | not reported | — | — |
| Humanity's Last Exam (Python + Search) | 29.8% ↗ | not reported | — | — |
| Humanity's Last Exam (thinking, text-only) | 15.9% ↗ | not reported | — | — |
| LiveCodeBench (2408-2505) (thinking) | 74.8% ↗ | not reported | — | — |
| SimpleBench (SimpleBench) | 40% ↗ | not reported | — | — |
| SimpleQA (search agent) | 93.4% ↗ | not reported | — | — |
| SWE Verified (Agent mode) | 66% ↗ | not reported | — | — |
| SWE-bench Multilingual (Agent mode) | 54.5% ↗ | not reported | — | — |
| Terminal-bench (Terminus 1 framework) | 31.3% ↗ | not reported | — | — |
| WeirdML (Håvard Tveit Ihle) | 38.4% ↗ | not reported | — | — |
| Only Qwen3.5 2B reports · 14 | ||||
| BFCL-V4 (thinking) | not reported | 43.6% ↗ | — | — |
| Chess Puzzles (Epoch AI run) | not reported | 0% ↗epoch run | — | — |
| GPQA (thinking) | not reported | 51.6% ↗ | — | — |
| HMMT Feb 25 (thinking) | not reported | 22.9% ↗ | — | — |
| IFBench (thinking) | not reported | 41.3% ↗ | — | — |
| IFEval (non-thinking) | not reported | 61.2% ↗ | — | — |
| IFEval (thinking) | not reported | 78.6% ↗ | — | — |
| LongBench v2 (thinking) | not reported | 38.7% ↗ | — | — |
| MMLU-Pro (non-thinking) | not reported | 55.3% ↗ | — | — |
| MMMLU (thinking) | not reported | 63.1% ↗ | — | — |
| MMMU (thinking) | not reported | 64.2% ↗ | — | — |
| MMMU-Pro (thinking) | not reported | 50.3% ↗ | — | — |
| SuperGPQA (thinking) | not reported | 37.5% ↗ | — | — |
| TAU2-Bench (thinking) | not reported | 48.8% ↗ | — | — |
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 V3.1 minus Qwen3.5 2B 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 V3.1 | Qwen3.5 2B | Edge |
|---|---|---|---|
| Context window | 128K tokens | 262K tokens | Qwen3.5 2B |
| Max output | — | — | — |
| Input price / 1M | — | — | — |
| Output price / 1M | — | — | — |
| Cached input / 1M | — | — | — |
| Input + output / 1M Lower is cheaper. List prices; batch, tool and regional fees excluded. | — | — | — |
| Modalities | text | text · vision · video | Qwen3.5 2B |
| Released | 2025-08-21 | 2026-03-02 | — |
| Cited benchmark scores | 18 | 16 | — |
Reliability
Provider status
Live from /statusMore matchups
DeepSeek V3.1 vs …
Models sharing the most benchmarksMore matchups
Qwen3.5 2B vs …
Models sharing the most benchmarksBuilt by Respan
Which one wins on your data?
Public benchmarks are a starting point. Run DeepSeek V3.1 and Qwen3.5 2B on your own prompts with Respan evals, or route to either through one gateway key with automatic failover.