Compare · ModelsLive · 3 picked · head to head
DeepSeek V3.2 Speciale vs gpt-oss-20b vs Qwen3.5 397B A17B
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
Qwen3.5 397B A17B wins on 18/24 benchmarks
Qwen3.5 397B A17B wins 18 of 24 shared benchmarks. Leads in speed · knowledge · coding.
Category leads
speed·Qwen3.5 397B A17Bmath·DeepSeek V3.2 Specialeknowledge·Qwen3.5 397B A17Blanguage·DeepSeek V3.2 Specialecoding·Qwen3.5 397B A17Barena·Qwen3.5 397B A17Bgeneral·Qwen3.5 397B A17B
Hype vs Reality
Attention vs performance
DeepSeek V3.2 Speciale
#6 by perf·no signal
gpt-oss-20b
#97 by perf·no signal
Qwen3.5 397B A17B
#48 by perf·#2 by attention
Best value
gpt-oss-20b
10.1x better value than DeepSeek V3.2 Speciale
DeepSeek V3.2 Speciale
97.8 pts/$
$0.80/M
gpt-oss-20b
983.3 pts/$
$0.05/M
Qwen3.5 397B A17B
29.6 pts/$
$2.02/M
Vendor risk
Mixed exposure
One or more vendors flagged
DeepSeek
$3.4B·Tier 1
OpenAI
$840.0B·Tier 1
Alibaba (Qwen)
$293.0B·Tier 1
Head to head
24 benchmarks · 3 models
DeepSeek V3.2 Specialegpt-oss-20bQwen3.5 397B A17B
Artificial Analysis · Quality Index
DeepSeek V3.2 Speciale leads by +8.0
DeepSeek V3.2 Speciale
29.4
gpt-oss-20b
9.0
Qwen3.5 397B A17B
21.4
OpenCompass · AIME2025
DeepSeek V3.2 Speciale leads by +3.7
DeepSeek V3.2 Speciale
96.0
gpt-oss-20b
87.9
Qwen3.5 397B A17B
92.3
OpenCompass · GPQA-Diamond
Qwen3.5 397B A17B leads by +1.7
DeepSeek V3.2 Speciale
86.7
gpt-oss-20b
68.9
Qwen3.5 397B A17B
88.4
OpenCompass · HLE
DeepSeek V3.2 Speciale leads by +1.1
DeepSeek V3.2 Speciale
28.6
gpt-oss-20b
11.6
Qwen3.5 397B A17B
27.5
OpenCompass · IFEval
DeepSeek V3.2 Speciale leads by +0.2
DeepSeek V3.2 Speciale
91.7
gpt-oss-20b
88.9
Qwen3.5 397B A17B
91.5
OpenCompass · LiveCodeBenchV6
Qwen3.5 397B A17B leads by +2.1
DeepSeek V3.2 Speciale
80.9
gpt-oss-20b
68.4
Qwen3.5 397B A17B
83.0
OpenCompass · MMLU-Pro
Qwen3.5 397B A17B leads by +2.1
DeepSeek V3.2 Speciale
85.5
gpt-oss-20b
72.8
Qwen3.5 397B A17B
87.6
Artificial Analysis · Agentic Index
Qwen3.5 397B A17B leads by +19.9
Artificial Analysis Agentic Index · a composite score measuring how well a model performs in agentic workflows · multi-step tool use, planning, error recovery, and autonomous task completion. Aggregates results from multiple agentic benchmarks including SWE-bench, tool-use tests, and planning evaluations. The canonical single-number metric for "how good is this model as an agent?"
DeepSeek V3.2 Speciale
0.0
Qwen3.5 397B A17B
19.9
Artificial Analysis · Coding Index
Qwen3.5 397B A17B leads by +10.3
Artificial Analysis Coding Index · a composite score that aggregates performance across multiple coding benchmarks into a single index. Tracks code generation quality, debugging ability, multi-language competence, and real-world software engineering tasks. Used by Artificial Analysis to rank model coding capability in a normalized, comparable format. Useful for developers choosing between models for coding-heavy workloads.
DeepSeek V3.2 Speciale
37.9
Qwen3.5 397B A17B
48.2
Artificial Analysis · CritPt
gpt-oss-20b leads by +0.5
gpt-oss-20b
1.4
Qwen3.5 397B A17B
0.9
Artificial Analysis · GDPval
Qwen3.5 397B A17B leads by +14.0
gpt-oss-20b
0.0
Qwen3.5 397B A17B
14.0
Artificial Analysis · GPQA Diamond
Qwen3.5 397B A17B leads by +17.3
gpt-oss-20b
68.8
Qwen3.5 397B A17B
86.1
Artificial Analysis · Humanity's Last Exam
Qwen3.5 397B A17B leads by +8.8
gpt-oss-20b
11.0
Qwen3.5 397B A17B
19.8
Artificial Analysis · IFBench
gpt-oss-20b leads by +13.5
gpt-oss-20b
65.1
Qwen3.5 397B A17B
51.6
Artificial Analysis · Long Context Reasoning
Qwen3.5 397B A17B leads by +29.6
gpt-oss-20b
34.7
Qwen3.5 397B A17B
64.3
Artificial Analysis · SciCode
Qwen3.5 397B A17B leads by +5.9
gpt-oss-20b
38.9
Qwen3.5 397B A17B
44.8
Artificial Analysis · tau2-Bench Telecom
Qwen3.5 397B A17B leads by +23.7
gpt-oss-20b
60.2
Qwen3.5 397B A17B
83.9
Artificial Analysis · Terminal-Bench Hard
Qwen3.5 397B A17B leads by +25.0
gpt-oss-20b
10.6
Qwen3.5 397B A17B
35.6
Chatbot Arena Elo · Overall
Qwen3.5 397B A17B leads by +124.2
gpt-oss-20b
1317.6
Qwen3.5 397B A17B
1441.8
Chess Puzzles
Qwen3.5 397B A17B leads by +8.5
Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities.
gpt-oss-20b
0.0
Qwen3.5 397B A17B
8.5
Dtbench
Qwen3.5 397B A17B leads by +32.5
gpt-oss-20b
46.7
Qwen3.5 397B A17B
79.1
GPQA diamond
Qwen3.5 397B A17B leads by +34.1
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
gpt-oss-20b
47.7
Qwen3.5 397B A17B
81.8
Lmca
Qwen3.5 397B A17B leads by +27.5
gpt-oss-20b
17.1
Qwen3.5 397B A17B
44.6
OTIS Mock AIME 2024-2025
Qwen3.5 397B A17B leads by +23.6
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
gpt-oss-20b
65.2
Qwen3.5 397B A17B
88.9
Full benchmark table
| Benchmark | DeepSeek V3.2 Speciale | gpt-oss-20b | Qwen3.5 397B A17B |
|---|---|---|---|
Artificial Analysis · Quality Index | 29.4 | 9.0 | 21.4 |
OpenCompass · AIME2025 | 96.0 | 87.9 | 92.3 |
OpenCompass · GPQA-Diamond | 86.7 | 68.9 | 88.4 |
OpenCompass · HLE | 28.6 | 11.6 | 27.5 |
OpenCompass · IFEval | 91.7 | 88.9 | 91.5 |
OpenCompass · LiveCodeBenchV6 | 80.9 | 68.4 | 83.0 |
OpenCompass · MMLU-Pro | 85.5 | 72.8 | 87.6 |
Artificial Analysis · Agentic Index Artificial Analysis Agentic Index · a composite score measuring how well a model performs in agentic workflows · multi-step tool use, planning, error recovery, and autonomous task completion. Aggregates results from multiple agentic benchmarks including SWE-bench, tool-use tests, and planning evaluations. The canonical single-number metric for "how good is this model as an agent?" | 0.0 | — | 19.9 |
Artificial Analysis · Coding Index Artificial Analysis Coding Index · a composite score that aggregates performance across multiple coding benchmarks into a single index. Tracks code generation quality, debugging ability, multi-language competence, and real-world software engineering tasks. Used by Artificial Analysis to rank model coding capability in a normalized, comparable format. Useful for developers choosing between models for coding-heavy workloads. | 37.9 | — | 48.2 |
Artificial Analysis · CritPt | — | 1.4 | 0.9 |
Artificial Analysis · GDPval | — | 0.0 | 14.0 |
Artificial Analysis · GPQA Diamond | — | 68.8 | 86.1 |
Artificial Analysis · Humanity's Last Exam | — | 11.0 | 19.8 |
Artificial Analysis · IFBench | — | 65.1 | 51.6 |
Artificial Analysis · Long Context Reasoning | — | 34.7 | 64.3 |
Artificial Analysis · SciCode | — | 38.9 | 44.8 |
Artificial Analysis · tau2-Bench Telecom | — | 60.2 | 83.9 |
Artificial Analysis · Terminal-Bench Hard | — | 10.6 | 35.6 |
Chatbot Arena Elo · Overall | — | 1317.6 | 1441.8 |
Chess Puzzles Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities. | — | 0.0 | 8.5 |
Dtbench | — | 46.7 | 79.1 |
GPQA diamond Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs. | — | 47.7 | 81.8 |
Lmca | — | 17.1 | 44.6 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | — | 65.2 | 88.9 |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| $0.40 | $1.20 | 164K tokens (~82 books) | $6.00 | |
| $0.02 | $0.09 | 131K tokens (~66 books) | $0.36 | |
| $0.55 | $3.50 | 262K tokens (~131 books) | $12.88 |
People also compared
DeepSeek V3.2 Speciale vs GPT-5.5 ProDeepSeek V3.2 Speciale vs GPT-5.5Claude Opus 5.5 vs DeepSeek V3.2 SpecialeDeepSeek V3.2 Speciale vs GPT-5 ChatClaude Mythos Preview vs DeepSeek V3.2 SpecialeDeepSeek V3.2 Speciale vs Step 3.5 FlashDeepSeek V3.2 Speciale vs GPT-6 AstraDeepSeek-V2 (MoE-236B, May 2024) vs DeepSeek V3.2 Speciale