Compare · ModelsLive · 3 picked · head to head
GPT-4 Turbo vs Qwen2.5 72B Instruct vs Qwen2-72B
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
Qwen2.5 72B Instruct wins on 13/21 benchmarks
Qwen2.5 72B Instruct wins 13 of 21 shared benchmarks. Leads in math · general · coding.
Category leads
knowledge·Qwen2-72Bmath·Qwen2.5 72B Instructgeneral·Qwen2.5 72B Instructcoding·Qwen2.5 72B Instructreasoning·Qwen2.5 72B Instructlanguage·Qwen2.5 72B Instructagentic·Qwen2.5 72B Instruct
Hype vs Reality
Attention vs performance
GPT-4 Turbo
#197 by perf·no signal
Qwen2.5 72B Instruct
#128 by perf·#2 by attention
Qwen2-72B
#203 by perf·#2 by attention
Best value
Qwen2.5 72B Instruct
66.4x better value than GPT-4 Turbo
GPT-4 Turbo
1.9 pts/$
$20.00/M
Qwen2.5 72B Instruct
129.5 pts/$
$0.38/M
Qwen2-72B
n/a
no price
Vendor risk
Who is behind the model
OpenAI
$840.0B·Tier 1
Alibaba (Qwen)
$293.0B·Tier 1
Alibaba (Qwen)
$293.0B·Tier 1
Head to head
21 benchmarks · 3 models
GPT-4 TurboQwen2.5 72B InstructQwen2-72B
CMMLU
Qwen2-72B leads by +4.0
GPT-4 Turbo
71.0
Qwen2.5 72B Instruct
85.7
Qwen2-72B
89.7
GPQA diamond
Qwen2.5 72B Instruct leads by +11.2
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
GPT-4 Turbo
7.5
Qwen2.5 72B Instruct
32.2
Qwen2-72B
21.0
MATH level 5
Qwen2.5 72B Instruct leads by +24.1
MATH Level 5 · the hardest tier of the MATH benchmark, featuring competition-level problems from AMC, AIME, and Olympiad-style mathematics.
GPT-4 Turbo
23.0
Qwen2.5 72B Instruct
63.2
Qwen2-72B
39.1
Metr Time Horizons
Qwen2.5 72B Instruct leads by +5.9
GPT-4 Turbo
29.3
Qwen2.5 72B Instruct
35.8
Qwen2-72B
29.9
MMLU
Qwen2.5 72B Instruct leads by +3.9
Massive Multitask Language Understanding · 57 subjects spanning STEM, humanities, social sciences, and more. The standard benchmark for broad knowledge.
GPT-4 Turbo
76.5
Qwen2.5 72B Instruct
80.4
Qwen2-72B
76.5
WeirdML
Qwen2.5 72B Instruct leads by +3.6
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
GPT-4 Turbo
12.4
Qwen2.5 72B Instruct
16.0
Qwen2-72B
11.3
Aider · Code Editing
Qwen2.5 72B Instruct leads by +9.8
Qwen2.5 72B Instruct
65.4
Qwen2-72B
55.6
BBH
Qwen2.5 72B Instruct leads by +6.2
BIG-Bench Hard · a curated subset of 23 challenging tasks from BIG-Bench where language models previously failed to outperform average humans.
GPT-4 Turbo
66.8
Qwen2.5 72B Instruct
73.1
Dtbench
Qwen2.5 72B Instruct leads by +0.4
GPT-4 Turbo
37.8
Qwen2.5 72B Instruct
38.2
HellaSwag
GPT-4 Turbo leads by +14.0
HellaSwag · tests commonsense reasoning by asking models to predict the most plausible continuation of everyday scenarios.
GPT-4 Turbo
93.7
Qwen2.5 72B Instruct
79.7
BBH (HuggingFace)
Qwen2.5 72B Instruct leads by +10.0
Qwen2.5 72B Instruct
61.9
Qwen2-72B
51.9
GPQA
Qwen2-72B leads by +2.6
Qwen2.5 72B Instruct
16.7
Qwen2-72B
19.2
IFEval
Qwen2.5 72B Instruct leads by +48.1
Qwen2.5 72B Instruct
86.4
Qwen2-72B
38.2
MATH Level 5
Qwen2.5 72B Instruct leads by +28.7
Qwen2.5 72B Instruct
59.8
Qwen2-72B
31.1
MMLU-PRO
Qwen2-72B leads by +1.2
Qwen2.5 72B Instruct
51.4
Qwen2-72B
52.6
MUSR
Qwen2-72B leads by +8.0
Qwen2.5 72B Instruct
11.7
Qwen2-72B
19.7
Lmca
GPT-4 Turbo leads by +4.3
GPT-4 Turbo
20.1
Qwen2.5 72B Instruct
15.8
OTIS Mock AIME 2024-2025
Qwen2.5 72B Instruct leads by +7.0
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
GPT-4 Turbo
1.0
Qwen2.5 72B Instruct
8.0
The Agent Company
Qwen2.5 72B Instruct leads by +4.6
The Agent Company · tests AI agents on realistic corporate tasks like email management, code review, data analysis, and cross-tool workflows.
Qwen2.5 72B Instruct
5.7
Qwen2-72B
1.1
TriviaQA
GPT-4 Turbo leads by +12.9
TriviaQA · reading comprehension benchmark with trivia questions, requiring models to find and reason over evidence from provided documents.
GPT-4 Turbo
84.8
Qwen2.5 72B Instruct
71.9
Winogrande
GPT-4 Turbo leads by +10.4
WinoGrande · large-scale commonsense reasoning benchmark where models must resolve ambiguous pronouns in carefully constructed sentence pairs.
GPT-4 Turbo
75.0
Qwen2.5 72B Instruct
64.6
Full benchmark table
| Benchmark | GPT-4 Turbo | Qwen2.5 72B Instruct | Qwen2-72B |
|---|---|---|---|
CMMLU | 71.0 | 85.7 | 89.7 |
GPQA diamond Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs. | 7.5 | 32.2 | 21.0 |
MATH level 5 MATH Level 5 · the hardest tier of the MATH benchmark, featuring competition-level problems from AMC, AIME, and Olympiad-style mathematics. | 23.0 | 63.2 | 39.1 |
Metr Time Horizons | 29.3 | 35.8 | 29.9 |
MMLU Massive Multitask Language Understanding · 57 subjects spanning STEM, humanities, social sciences, and more. The standard benchmark for broad knowledge. | 76.5 | 80.4 | 76.5 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 12.4 | 16.0 | 11.3 |
Aider · Code Editing | — | 65.4 | 55.6 |
BBH BIG-Bench Hard · a curated subset of 23 challenging tasks from BIG-Bench where language models previously failed to outperform average humans. | 66.8 | 73.1 | — |
Dtbench | 37.8 | 38.2 | — |
HellaSwag HellaSwag · tests commonsense reasoning by asking models to predict the most plausible continuation of everyday scenarios. | 93.7 | 79.7 | — |
BBH (HuggingFace) | — | 61.9 | 51.9 |
GPQA | — | 16.7 | 19.2 |
IFEval | — | 86.4 | 38.2 |
MATH Level 5 | — | 59.8 | 31.1 |
MMLU-PRO | — | 51.4 | 52.6 |
MUSR | — | 11.7 | 19.7 |
Lmca | 20.1 | 15.8 | — |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 1.0 | 8.0 | — |
The Agent Company The Agent Company · tests AI agents on realistic corporate tasks like email management, code review, data analysis, and cross-tool workflows. | — | 5.7 | 1.1 |
TriviaQA TriviaQA · reading comprehension benchmark with trivia questions, requiring models to find and reason over evidence from provided documents. | 84.8 | 71.9 | — |
Winogrande WinoGrande · large-scale commonsense reasoning benchmark where models must resolve ambiguous pronouns in carefully constructed sentence pairs. | 75.0 | 64.6 | — |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| $10.00 | $30.00 | 128K tokens (~64 books) | $150.00 | |
| $0.36 | $0.40 | 33K tokens (~16 books) | $3.70 | |
| — | — | — | — |