Compare · ModelsLive · 3 picked · head to head
Gemma 4 31B vs GLM 5.1 vs Qwen3.7 Max
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
Qwen3.7 Max wins on 18/24 benchmarks
Qwen3.7 Max wins 18 of 24 shared benchmarks. Leads in speed · knowledge · reasoning.
Category leads
speed·Qwen3.7 Maxknowledge·Qwen3.7 Maxcoding·GLM 5.1reasoning·Qwen3.7 Maxlanguage·Qwen3.7 Maxmath·Qwen3.7 Maxarena·GLM 5.1general·Qwen3.7 Max
Hype vs Reality
Attention vs performance
Gemma 4 31B
#101 by perf·#7 by attention
GLM 5.1
#84 by perf·#3 by attention
Qwen3.7 Max
#41 by perf·#2 by attention
Best value
Gemma 4 31B
9.0x better value than GLM 5.1
Gemma 4 31B
245.6 pts/$
$0.22/M
GLM 5.1
27.1 pts/$
$2.00/M
Qwen3.7 Max
24.5 pts/$
$2.50/M
Vendor risk
Who is behind the model
Google DeepMind
$4.20T·Tier 1
z-ai
private · undisclosed
Alibaba (Qwen)
$293.0B·Tier 1
Head to head
24 benchmarks · 3 models
Gemma 4 31BGLM 5.1Qwen3.7 Max
Artificial Analysis · Quality Index
Qwen3.7 Max leads by +5.8
Gemma 4 31B
14.7
GLM 5.1
40.2
Qwen3.7 Max
46.0
Chess Puzzles
Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities.
Gemma 4 31B
0.0
GLM 5.1
14.8
Qwen3.7 Max
14.8
GPQA diamond
Qwen3.7 Max leads by +1.3
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
Gemma 4 31B
67.7
GLM 5.1
86.5
Qwen3.7 Max
87.9
LiveBench · Agentic Coding
GLM 5.1 leads by +3.3
Gemma 4 31B
40.0
GLM 5.1
55.0
Qwen3.7 Max
51.7
LiveBench · Coding
GLM 5.1 leads by +1.2
Gemma 4 31B
60.3
GLM 5.1
75.4
Qwen3.7 Max
74.2
LiveBench · Data Analysis
Qwen3.7 Max leads by +8.6
Gemma 4 31B
58.8
GLM 5.1
63.2
Qwen3.7 Max
71.8
LiveBench · If
Qwen3.7 Max leads by +5.6
Gemma 4 31B
67.6
GLM 5.1
68.5
Qwen3.7 Max
74.0
LiveBench · Language
Qwen3.7 Max leads by +8.0
Gemma 4 31B
71.3
GLM 5.1
71.8
Qwen3.7 Max
79.7
LiveBench · Mathematics
Qwen3.7 Max leads by +0.4
Gemma 4 31B
73.9
GLM 5.1
84.9
Qwen3.7 Max
85.3
LiveBench · Overall
Qwen3.7 Max leads by +4.1
Gemma 4 31B
61.6
GLM 5.1
70.2
Qwen3.7 Max
74.3
LiveBench · Reasoning
Qwen3.7 Max leads by +10.8
Gemma 4 31B
59.4
GLM 5.1
72.5
Qwen3.7 Max
83.3
OTIS Mock AIME 2024-2025
Qwen3.7 Max leads by +2.2
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
Gemma 4 31B
73.3
GLM 5.1
93.3
Qwen3.7 Max
95.5
SimpleQA Verified
Qwen3.7 Max leads by +21.8
SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information.
Gemma 4 31B
10.4
GLM 5.1
34.0
Qwen3.7 Max
55.8
Artificial Analysis · Agentic Index
Qwen3.7 Max leads by +0.7
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?"
GLM 5.1
29.9
Qwen3.7 Max
30.6
Artificial Analysis · Coding Index
Qwen3.7 Max leads by +10.2
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.
GLM 5.1
55.8
Qwen3.7 Max
66.0
Chatbot Arena Elo · Coding
GLM 5.1 leads by +143.8
Gemma 4 31B
1364.7
GLM 5.1
1508.5
Chatbot Arena Elo · Overall
GLM 5.1 leads by +11.7
Gemma 4 31B
1452.8
GLM 5.1
1464.6
Dtbench
Qwen3.7 Max leads by +16.0
Gemma 4 31B
71.1
Qwen3.7 Max
87.1
FrontierMath-Tiers-1-3-v2-Private
Qwen3.7 Max leads by +27.7
GLM 5.1
36.8
Qwen3.7 Max
64.6
Lmca
Qwen3.7 Max leads by +5.6
Gemma 4 31B
46.2
Qwen3.7 Max
51.8
Proofbench
Qwen3.7 Max leads by +3.8
GLM 5.1
22.2
Qwen3.7 Max
26.0
SimpleBench
Qwen3.7 Max leads by +18.4
SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking.
GLM 5.1
46.1
Qwen3.7 Max
64.5
SWE-Bench verified
Qwen3.7 Max leads by +3.1
SWE-bench Verified · 500 human-validated tasks from 12 real Python repositories (Django, Flask, scikit-learn, sympy, and others). Each task requires the model to produce a git patch that resolves a real GitHub issue and passes the test suite. The verified subset eliminates ambiguous tasks from the original SWE-bench. Claude Mythos Preview leads at 93.9%, crossing 90% for the first time in 2026. Opus 4.6 scores 80.8%. The benchmark remains the most-cited evaluation for code-generation capability.
GLM 5.1
74.2
Qwen3.7 Max
77.3
WeirdML
GLM 5.1 leads by +4.8
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
Gemma 4 31B
52.3
GLM 5.1
57.1
Full benchmark table
| Benchmark | Gemma 4 31B | GLM 5.1 | Qwen3.7 Max |
|---|---|---|---|
Artificial Analysis · Quality Index | 14.7 | 40.2 | 46.0 |
Chess Puzzles Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities. | 0.0 | 14.8 | 14.8 |
GPQA diamond Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs. | 67.7 | 86.5 | 87.9 |
LiveBench · Agentic Coding | 40.0 | 55.0 | 51.7 |
LiveBench · Coding | 60.3 | 75.4 | 74.2 |
LiveBench · Data Analysis | 58.8 | 63.2 | 71.8 |
LiveBench · If | 67.6 | 68.5 | 74.0 |
LiveBench · Language | 71.3 | 71.8 | 79.7 |
LiveBench · Mathematics | 73.9 | 84.9 | 85.3 |
LiveBench · Overall | 61.6 | 70.2 | 74.3 |
LiveBench · Reasoning | 59.4 | 72.5 | 83.3 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 73.3 | 93.3 | 95.5 |
SimpleQA Verified SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information. | 10.4 | 34.0 | 55.8 |
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?" | — | 29.9 | 30.6 |
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. | — | 55.8 | 66.0 |
Chatbot Arena Elo · Coding | 1364.7 | 1508.5 | — |
Chatbot Arena Elo · Overall | 1452.8 | 1464.6 | — |
Dtbench | 71.1 | — | 87.1 |
FrontierMath-Tiers-1-3-v2-Private | — | 36.8 | 64.6 |
Lmca | 46.2 | — | 51.8 |
Proofbench | — | 22.2 | 26.0 |
SimpleBench SimpleBench · tests fundamental reasoning capabilities with straightforward problems designed to expose gaps in basic logical and spatial thinking. | — | 46.1 | 64.5 |
SWE-Bench verified SWE-bench Verified · 500 human-validated tasks from 12 real Python repositories (Django, Flask, scikit-learn, sympy, and others). Each task requires the model to produce a git patch that resolves a real GitHub issue and passes the test suite. The verified subset eliminates ambiguous tasks from the original SWE-bench. Claude Mythos Preview leads at 93.9%, crossing 90% for the first time in 2026. Opus 4.6 scores 80.8%. The benchmark remains the most-cited evaluation for code-generation capability. | — | 74.2 | 77.3 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 52.3 | 57.1 | — |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| $0.09 | $0.34 | 262K tokens (~131 books) | $1.53 | |
| $0.97 | $3.04 | 205K tokens (~102 books) | $14.83 | |
| $1.25 | $3.75 | 1.0M tokens (~500 books) | $18.75 |