Model comparison
GLM-5.2 vs Qwen3.5 397B
Head-to-head evidence from 27 shared benchmark results across 6 categories. Overall scores shown here use the public BenchAlign v5 ranking lane.
Public leaderboard positions: GLM-5.2 #37 (Estimated); Qwen3.5 397B #71 (Estimated). Intervals and evidence labels describe ranking uncertainty, not a guarantee for a specific workload.
Evidence parity. GLM-5.2 and Qwen3.5 397B share 27 comparable benchmark results. 4 of 8 categories are comparable. 16 results are unique to GLM-5.2; 28 to Qwen3.5 397B.
Updated July 23, 2026- Shared results
- 27
- GLM-5.2 only
- 16
- Qwen3.5 397B only
- 28
- Comparable categories
- 4 / 8
Pick GLM-5.2 if you want the stronger benchmark profile. Qwen3.5 397B 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 27 shared benchmark results across 6 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-5.2 is clearly ahead on the BenchAlign aggregate, 63.96 to 57.01. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.
GLM-5.2's sharpest advantage is in agentic, where it averages 81 against 56.5. The single biggest benchmark swing on the page is Terminal-Bench 2.0, 81% to 52.5%. Qwen3.5 397B does hit back in coding, so the answer changes if that is the part of the workload you care about most.
GLM-5.2 is also the more expensive model on tokens at $1.40 input / $4.40 output per 1M tokens, versus $0.60 input / $3.60 output per 1M tokens for Qwen3.5 397B. GLM-5.2 is the reasoning model in the pair, while Qwen3.5 397B 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. GLM-5.2 gives you the larger context window at 1M, compared with 128K for Qwen3.5 397B.
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 | GLM-5.2 | Δ | Qwen3.5 397B |
|---|---|---|---|
| Agentic | GLM-5.281.0 | Margin← 24.5 | Qwen3.5 397B56.5 |
| Math | GLM-5.295.9 | Margin← 5.3 | Qwen3.5 397B90.6 |
| Coding | GLM-5.262.1 | Margin→ 4.4 | Qwen3.5 397B66.5 |
| Knowledge | GLM-5.259.6 | Margin← 3.0 | Qwen3.5 397B56.6 |
| Reasoning | GLM-5.2Not measured | MarginNo overlap | Qwen3.5 397B63.2 |
| Multilingual | GLM-5.2Not measured | MarginNo overlap | Qwen3.5 397B84.7 |
| Multimodal | GLM-5.2Not measured | MarginNo overlap | Qwen3.5 397B79.6 |
| Inst. Following | GLM-5.2Not measured | MarginNo overlap | Qwen3.5 397B92.6 |
Decisive benchmark drivers
The largest measured benchmark gaps in this matchup, with exact reported values.
More
- Source ↗
Terminal-Bench 2.0
AgenticA 81%B 52.5%Winner: GLM-5.2Δ 28.5Terminal-Bench 2.0: GLM-5.2 scored 81%; Qwen3.5 397B scored 52.5%. GLM-5.2 wins this benchmark. - Source ↗
HLE
KnowledgeA 54.7%B 28.7%Winner: GLM-5.2Δ 26HLE: GLM-5.2 scored 54.7%; Qwen3.5 397B scored 28.7%. GLM-5.2 wins this benchmark. - Source ↗
SWE-bench Pro
CodingA 62.1%B 50.9%Winner: GLM-5.2Δ 11.2SWE-bench Pro: GLM-5.2 scored 62.1%; Qwen3.5 397B scored 50.9%. GLM-5.2 wins this benchmark. - Source ↗
AIME26
MathA 99.2%B 93.3%Winner: GLM-5.2Δ 5.9AIME26: GLM-5.2 scored 99.2%; Qwen3.5 397B scored 93.3%. GLM-5.2 wins this benchmark. - Source ↗
HMMT Feb 2026
MathA 92.5%B 87.9%Winner: GLM-5.2Δ 4.6HMMT Feb 2026: GLM-5.2 scored 92.5%; Qwen3.5 397B scored 87.9%. GLM-5.2 wins this benchmark.
Operational comparison
Runtime and commercial metrics are compared only when both models have a complete sourced value.
| Metric | GLM-5.2 | Qwen3.5 397B | Comparison |
|---|---|---|---|
| Input / output priceUSD per 1M tokens | GLM-5.2$1.4 input / $4.4 output | Qwen3.5 397B$0.6 input / $3.6 output | Qwen3.5 397B has the lower combined listed price. |
| Generation speedtokens per second | GLM-5.2Not available | Qwen3.5 397B96 tok/s | A complete speed comparison is not available. |
| First-answer latencyseconds to first token | GLM-5.2Not available | Qwen3.5 397B2.44 s | A complete latency comparison is not available. |
| Context windowmaximum listed tokens | GLM-5.21M | Qwen3.5 397B128K | GLM-5.2 lists the larger context window. |
Benchmark Deep Dive
AgenticGLM-5.2 wins26 benchmarks
| Benchmark | GLM-5.2 | Qwen3.5 397B | Result |
|---|---|---|---|
| Terminal-Bench 2.0Source | 81% | 52.5% | GLM-5.2 leads |
| MCP AtlasSource | 76.8% | 46.1% | GLM-5.2 leads |
| ToolathlonSource | 48.2% | 36.3% | GLM-5.2 leads |
| AA Agentic IndexSource | 43.1% | 19.9% | GLM-5.2 leads |
| τ²-bench resultsSource | 99.1% | 95.6% | GLM-5.2 leads |
| GDPval-AASource | 50.7% | 23.1% | GLM-5.2 leads |
| GDPval-AASource | 1514 | 962 | GLM-5.2 leads |
| APEX-Agents-AASource | 33.7% | 15.3% | GLM-5.2 leads |
| ResearchClawBenchSource | 20.7% | 14.2% | GLM-5.2 leads |
| AA BriefcaseSource | 1260 | — | Not comparable |
| AA AutomationBenchSource | 27.8% | — | Not comparable |
| AA EnterpriseOps-GymSource | 42.7% | — | Not comparable |
| AA Harvey LABSource | 91.0% | — | Not comparable |
| AA ITBenchSource | 42.7% | — | Not comparable |
| AA Tau3 BankingSource | 26.8% | — | Not comparable |
| terminalBenchHardSource | 50.8% | — | Not comparable |
| aaTerminalBench21Source | 77.9% | — | Not comparable |
| BrowseCompSource | — | 62% | Not comparable |
| Claw-EvalSource | — | 56.8% | Not comparable |
| QwenClawBenchSource | — | 51.8% | Not comparable |
| τ³-bench resultsSource | — | 68.4% | Not comparable |
| VITA-BenchSource | — | 43.7% | Not comparable |
| DeepPlanningSource | — | 37.6% | Not comparable |
| MCP-TasksSource | — | 74.2% | Not comparable |
| WideResearchSource | — | 74.0% | Not comparable |
| Gert LabsSource | — | 46.76% | Not comparable |
CodingQwen3.5 397B wins9 benchmarks
| Benchmark | GLM-5.2 | Qwen3.5 397B | Result |
|---|---|---|---|
| SWE-bench ProSource | 62.1% | 50.9% | GLM-5.2 leads |
| NL2RepoSource | 48.9% | — | Not comparable |
| Terminal-Bench 2.0Source | 81.0% | — | Not comparable |
| ProgramBenchSource | 63.7% | — | Not comparable |
| cursorBench32Source | 55.0% | — | Not comparable |
| AA Coding IndexSource | 68.8% | 48.2% | GLM-5.2 leads |
| AA-SciCodeSource | 50.5% | 42.0% | GLM-5.2 leads |
| SWE-bench VerifiedSource | — | 76.2% | Not comparable |
| LiveCodeBench v6Source | — | 83.6% | Not comparable |
Reasoning4 benchmarks
KnowledgeGLM-5.2 wins15 benchmarks
| Benchmark | GLM-5.2 | Qwen3.5 397B | Result |
|---|---|---|---|
| GPQASource | 91.2% | 88.4% | GLM-5.2 leads |
| GPQA-DSource | 91.2% | — | Not comparable |
| HLESource | 54.7% | 28.7% | GLM-5.2 leads |
| HLE w/o toolsSource | 40.5% | — | Not comparable |
| Artificial Analysis Intelligence IndexSource | 51.1% | 33.7% | GLM-5.2 leads |
| AA-GPQA DiamondSource | 89.5% | 89.3% | GLM-5.2 leads |
| AA-HLESource | 40.1% | 27.3% | GLM-5.2 leads |
| AA-Omniscience IndexSource | 4.0% | -29.8% | GLM-5.2 leads |
| AA-Omniscience AccuracySource | 25.1% | 31.4% | Qwen3.5 397B leads |
| AA-Omniscience Hallucination RateSource | 28.1% | 89.1% | GLM-5.2 leads |
| AA Openness IndexSource | 44.4% | — | Not comparable |
| SuperGPQASource | — | 70.4% | Not comparable |
| MMLU-ProSource | — | 87.8% | Not comparable |
| MMLU-ReduxSource | — | 94.9% | Not comparable |
| C-EvalSource | — | 93% | Not comparable |
MathGLM-5.2 wins5 benchmarks
Multilingual2 benchmarks
Multimodal8 benchmarks
| Benchmark | GLM-5.2 | Qwen3.5 397B | Result |
|---|---|---|---|
| Design Arena WebsiteSource | 1340 | — | Not comparable |
| MMMU-ProSource | — | 79% | Not comparable |
| MathVisionSource | — | 88.6% | Not comparable |
| CharXivSource | — | 80.8% | Not comparable |
| VideoMMMUSource | — | 84.7% | Not comparable |
| ScreenSpot ProSource | — | 65.6% | Not comparable |
| V*Source | — | 95.8% | Not comparable |
| AA-MMMU-ProSource | — | 77.3% | Not comparable |
Frequently Asked Questions (5)
Which is better, GLM-5.2 or Qwen3.5 397B?
GLM-5.2 is ahead on BenchLM's BenchAlign leaderboard, 63.96 to 57.01. The biggest single separator in this matchup is Terminal-Bench 2.0, where the scores are 81% and 52.5%.
Which is better for knowledge tasks, GLM-5.2 or Qwen3.5 397B?
GLM-5.2 has the edge for knowledge tasks in this comparison, averaging 59.6 versus 56.6. Inside this category, AA-Omniscience Hallucination Rate is the benchmark that creates the most daylight between them.
Which is better for coding, GLM-5.2 or Qwen3.5 397B?
Qwen3.5 397B has the edge for coding in this comparison, averaging 66.5 versus 62.1. Inside this category, AA Coding Index is the benchmark that creates the most daylight between them.
Which is better for math, GLM-5.2 or Qwen3.5 397B?
GLM-5.2 has the edge for math in this comparison, averaging 95.9 versus 90.6. Inside this category, MMAnswerBench is the benchmark that creates the most daylight between them.
Which is better for agentic tasks, GLM-5.2 or Qwen3.5 397B?
GLM-5.2 has the edge for agentic tasks in this comparison, averaging 81 versus 56.5. Inside this category, GDPval-AA is the benchmark that creates the most daylight between them.
Related Comparisons
Explore More
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.