Skip to main content

Model comparison

GLM-4.7 vs GLM-5

Data verified

Head-to-head evidence from 21 shared benchmark results across 7 categories. Overall scores shown here use the public BenchAlign v5 ranking lane.

61.16/100
Margin
4.9pts
winning →
Z.AI
66.06/100
1 category wins3 category wins

Public leaderboard positions: GLM-4.7 #42 (Supported); GLM-5 #28 (Supported). Intervals and evidence labels describe ranking uncertainty, not a guarantee for a specific workload.

Evidence parity. GLM-4.7 and GLM-5 share 21 comparable benchmark results. 4 of 8 categories are comparable. 9 results are unique to GLM-4.7; 28 to GLM-5.

Updated July 20, 2026
Shared results
21
GLM-4.7 only
9
GLM-5 only
28
Comparable categories
4 / 8

Pick GLM-5 if you want the stronger benchmark profile. GLM-4.7 only becomes the better choice if coding is the priority or you want the cheaper token bill.

Confidence note. This is a partial-evidence comparison with 21 shared benchmark results across 7 evidence categories; 4 of 8 categories currently have scoreable aggregates for both models. Treat the verdict as directional until coverage is more balanced.

Why this result

GLM-5 is clearly ahead on the BenchAlign aggregate, 66.06 to 61.16. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.

GLM-5's sharpest advantage is in mathematics, where it averages 56.3 against 1.8. The single biggest benchmark swing on the page is HLE, 24.8% to 50.4%. GLM-4.7 does hit back in coding, so the answer changes if that is the part of the workload you care about most.

GLM-5 is also the more expensive model on tokens at $1.00 input / $3.20 output per 1M tokens, versus $0.00 input / $0.00 output per 1M tokens for GLM-4.7. That is roughly Infinityx on output cost alone. GLM-4.7 is the reasoning model in the pair, while GLM-5 is not. That usually helps on harder chain-of-thought-heavy tests, but it can also mean more latency and more token spend in real use.

Category breakdown

Exact category averages are shown below. Not measured means BenchLM does not have enough sourced public coverage for that model and category.

Category scores and score margins for GLM-4.7 and GLM-5
CategoryGLM-4.7ΔGLM-5
MathGLM-4.71.8Margin 54.5GLM-556.3
KnowledgeGLM-4.751.8Margin 14.6GLM-566.4
AgenticGLM-4.745.7Margin 10.5GLM-556.2
CodingGLM-4.775.4Margin 9.1GLM-566.3
ReasoningGLM-4.7Not measuredMarginNo overlapGLM-560.8
MultilingualGLM-4.7Not measuredMarginNo overlapGLM-583.1
Inst. FollowingGLM-4.7Not measuredMarginNo overlapGLM-592.6

Decisive benchmark drivers

The largest measured benchmark gaps in this matchup, with exact reported values.

More
A · GLM-4.7B · GLM-5
  1. HLE

    Knowledge
    Source ↗
    A 24.8%B 50.4%
    Winner: GLM-5Δ 25.6
    HLE: GLM-4.7 scored 24.8%; GLM-5 scored 50.4%. GLM-5 wins this benchmark.
  2. Terminal-Bench 2.0

    Agentic
    Source ↗
    A 41%B 56.2%
    Winner: GLM-5Δ 15.2
    Terminal-Bench 2.0: GLM-4.7 scored 41%; GLM-5 scored 56.2%. GLM-5 wins this benchmark.
  3. FrontierMath v2 (Tiers 1-3)

    Math
    Source ↗
    A 2.439%B 16.434%
    Winner: GLM-5Δ 14
    FrontierMath v2 (Tiers 1-3): GLM-4.7 scored 2.439%; GLM-5 scored 16.434%. GLM-5 wins this benchmark.
  4. SWE-Rebench

    Coding
    Source ↗
    A 58.7%B 62.8%
    Winner: GLM-5Δ 4.1
    SWE-Rebench: GLM-4.7 scored 58.7%; GLM-5 scored 62.8%. GLM-5 wins this benchmark.
  5. SWE-bench Verified

    Coding
    Source ↗
    A 73.8%B 77.8%
    Winner: GLM-5Δ 4
    SWE-bench Verified: GLM-4.7 scored 73.8%; GLM-5 scored 77.8%. GLM-5 wins this benchmark.

Operational comparison

Runtime and commercial metrics are compared only when both models have a complete sourced value.

MetricGLM-4.7GLM-5Comparison
Input / output priceUSD per 1M tokensGLM-4.7$0 input / $0 outputGLM-5$1 input / $3.2 outputGLM-4.7 has the lower combined listed price.
Generation speedtokens per secondGLM-4.782 tok/sGLM-574 tok/sGLM-4.7 has the higher measured throughput.
First-answer latencyseconds to first tokenGLM-4.71.10 sGLM-51.64 sGLM-4.7 reaches the first token sooner.
Context windowmaximum listed tokensGLM-4.7200KGLM-5200KListed context windows are equal.

Benchmark Deep Dive

AgenticGLM-5 wins
BenchmarkGLM-4.7GLM-5Result
Terminal-Bench 2.0Source 41%56.2%GLM-5 leads
BrowseCompSource 52%Not comparable
VITA-BenchSource 15.5%Not comparable
AA Agentic IndexSource 25.4%Not comparable
τ²-bench resultsSource 95.9%98.2%GLM-5 leads
Gert LabsSource 39.95%50.99%GLM-5 leads
GDPval-AASource 33.3%Not comparable
GDPval-AASource 1165Not comparable
Claw-EvalSource 57.7%Not comparable
QwenClawBenchSource 54.1%Not comparable
τ³-bench resultsSource 65.6%Not comparable
DeepPlanningSource 14.6%Not comparable
ToolathlonSource 38%Not comparable
MCP AtlasSource 31.1%Not comparable
MCP-TasksSource 60.8%Not comparable
WideResearchSource 69.8%Not comparable
CyberGymSource 43.2%Not comparable
APEX-Agents-AASource 14.5%Not comparable
CodingGLM-4.7 wins
BenchmarkGLM-4.7GLM-5Result
SWE-bench VerifiedSource 73.8%77.8%GLM-5 leads
LiveCodeBenchSource 84.9%Not comparable
SWE-RebenchSource 58.7%62.8%GLM-5 leads
AA Coding IndexSource 45.3%Not comparable
AA-SciCodeSource 45.1%46.2%GLM-5 leads
AA LiveCodeBenchSource 89.4%Not comparable
SWE-bench Verified*Source 72.8%Not comparable
SWE-bench ProSource 55.1%Not comparable
SWE MultilingualSource 73.3%Not comparable
React Native EvalsSource 74.8%Not comparable
Reasoning
BenchmarkGLM-4.7GLM-5Result
AA-LCRSource 64.0%63.3%GLM-4.7 leads
CritPtSource 1.7%2.0%GLM-5 leads
LongBench v2Source 60.8%Not comparable
AI-NeedleSource 63.3%Not comparable
KnowledgeGLM-5 wins
BenchmarkGLM-4.7GLM-5Result
GPQASource 85.7%86%GLM-5 leads
MMLU-ProSource 84.3%85.7%GLM-5 leads
HLESource 24.8%50.4%GLM-5 leads
Artificial Analysis Intelligence IndexSource 33.7%39.5%GLM-5 leads
AA-GPQA DiamondSource 85.9%82.0%GLM-4.7 leads
AA-HLESource 25.1%27.2%GLM-5 leads
AA-Omniscience IndexSource -34.6%2.0%GLM-5 leads
AA-Omniscience AccuracySource 29.3%26.9%GLM-4.7 leads
AA-Omniscience Hallucination RateSource 90.3%34.0%GLM-5 leads
GPQA-DSource 86.0%Not comparable
SuperGPQASource 66.8%Not comparable
MMLU-Pro (Arcee)Source 85.8%Not comparable
MathGLM-5 wins
BenchmarkGLM-4.7GLM-5Result
AIME 2025Source 95.7%Not comparable
FrontierMath v2 (Tiers 1-3)Source 2.439%16.434%GLM-5 leads
FrontierMath v2 (Tier 4)Source 0.000%2.100%GLM-5 leads
AIME26Source 95.8%Not comparable
AIME25 (Arcee)Source 93.3%Not comparable
HMMT Feb 2025Source 97.5%Not comparable
HMMT Nov 2025Source 96.9%Not comparable
HMMT Feb 2026Source 86.4%Not comparable
MMAnswerBenchSource 82.5%Not comparable
Multilingual
BenchmarkGLM-4.7GLM-5Result
MMLU-ProXSource 83.1%Not comparable
NOVA-63Source 55.1%Not comparable
Multimodal
BenchmarkGLM-4.7GLM-5Result
Design Arena WebsiteSource 12581280GLM-5 leads
Inst. Following
BenchmarkGLM-4.7GLM-5Result
AA-IFBenchSource 67.9%72.3%GLM-5 leads
IFEvalSource 92.6%Not comparable
Frequently Asked Questions (5)

Which is better, GLM-4.7 or GLM-5?

GLM-5 is ahead on BenchLM's BenchAlign leaderboard, 66.06 to 61.16. The biggest single separator in this matchup is HLE, where the scores are 24.8% and 50.4%.

Which is better for knowledge tasks, GLM-4.7 or GLM-5?

GLM-5 has the edge for knowledge tasks in this comparison, averaging 66.4 versus 51.8. Inside this category, AA-Omniscience Hallucination Rate is the benchmark that creates the most daylight between them.

Which is better for coding, GLM-4.7 or GLM-5?

GLM-4.7 has the edge for coding in this comparison, averaging 75.4 versus 66.3. Inside this category, SWE-Rebench is the benchmark that creates the most daylight between them.

Which is better for math, GLM-4.7 or GLM-5?

GLM-5 has the edge for math in this comparison, averaging 56.3 versus 1.8. Inside this category, FrontierMath v2 (Tiers 1-3) is the benchmark that creates the most daylight between them.

Which is better for agentic tasks, GLM-4.7 or GLM-5?

GLM-5 has the edge for agentic tasks in this comparison, averaging 56.2 versus 45.7. Inside this category, Terminal-Bench 2.0 is the benchmark that creates the most daylight between them.

Related Comparisons

Last updated: July 20, 2026

Choose a model with this week’s evidence

Join 2,000+ readers for ranking moves, pricing changes, and the claims that still need proof.

One email each week. Unsubscribe anytime.