Model comparison
GLM-5.2 vs GPT-5.4
GPT-5.4 is the stronger model overall, scoring 59.4 to 51.1 on the Noometry Index. GLM-5.2 costs 2.6× less per token, which makes it the better buy when GPT-5.4's lead doesn't matter for your workload.
Last verified . 48 shared benchmarks.
Summary
- They share 48 benchmarks with published results for both. GLM-5.2 scores higher in 0 categories and GPT-5.4 in 9 categories; 7 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where GPT-5.4 leads 61.8 to 42.3.
- The biggest single-benchmark swing is ARC-AGI-2: 22.8% for GLM-5.2 and 74% for GPT-5.4.
- GLM-5.2 is cheaper at $1.40 / $4.40 per million input/output tokens, against $2.50 / $15 for GPT-5.4.
- GPT-5.4 accepts more context: 1.05M tokens versus 1M.
- GLM-5.2 has downloadable open weights; the other is API-only.
Side by side
| GLM-5.2 | GPT-5.4 | |
|---|---|---|
| Provider | Z.ai (Zhipu) | OpenAI |
| Noometry Index | 51.1 | 59.4 |
| Released | 2026-06-13 | 2026-03-05 |
| Weights | Open | Proprietary |
| Context window | 1M | 1.05M |
| Max output | 131K | 128K |
| Input $ / M tokens | $1.40 | $2.50 |
| Output $ / M tokens | $4.40 | $15 |
| Results tracked | 51 | 68 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-5.4 leads
GLM-5.2: 51.3 (#41), GPT-5.4: 52.6 (#33)
| Benchmark | GLM-5.2 | GPT-5.4 |
|---|---|---|
| SWE-bench Verified | 78.7% | 76.9% |
| DeepSWE | 43.8% | 51.8% |
| LMArena WebDev | 1603 | 1465 |
| SciCode | 50.5% | 56.6% |
| WeirdML | 70.1% | 77.7% |
| LMArena Coding | 1485 | 1497 |
| ALE-Bench | 1,047 | 1,607 |
| FrontierCode | 24.5% | — |
| GSO | — | 31.4% |
| MirrorCode | — | 15.6% |
| AlgoTune | — | 1.85 |
Agentic & Tool Use GPT-5.4 leads
GLM-5.2: 32.4 (#63), GPT-5.4: 46.5 (#13)
| Benchmark | GLM-5.2 | GPT-5.4 |
|---|---|---|
| APEX-Agents | 45.2% | 52.4% |
| τ²-bench Banking | 37.1% | 39.4% |
| PostTrainBench | 31.7% | 19% |
| GBAEval | 0% | 45.1% |
| Vending-Bench 2 | 8,314 | 6,144 |
| Terminal-Bench | — | 81.8% |
| DeepResearch Bench | — | 35.1% |
| LMArena Search | — | 1197 |
| METR Time Horizons | — | 74.3% |
Reasoning GPT-5.4 leads
GLM-5.2: 42.3 (#52), GPT-5.4: 61.8 (#19)
| Benchmark | GLM-5.2 | GPT-5.4 |
|---|---|---|
| ARC-AGI-2 | 22.8% | 74% |
| Kagi LLM Benchmark | 62.6% | 63.8% |
| NYT Connections (extended) | 74.3% | 91.3% |
| ARC-AGI-1 | 77% | 93.7% |
| CritPt | 20.9% | 23.4% |
| Chess Puzzles | 21% | 44% |
| EBR-Bench | 9.5% | 25.4% |
| LMArena Hard Prompts | 1480 | 1485 |
| Mystery Game Puzzles | 19% | 37% |
| DTBench | 93.6% | 94.4% |
| LMCA | 45.8% | 52% |
| Epoch Capabilities Index | 151.78 | 156.81 |
| SimpleBench | 58.8% | — |
| EnigmaEval | — | 16% |
| Thematic Generalization | — | 80% |
| Surface Evolver Bench | 55.6% | — |
| ForecastBench | — | 59.5 |
Math GPT-5.4 leads
GLM-5.2: 55.7 (#43), GPT-5.4: 73.5 (#19)
| Benchmark | GLM-5.2 | GPT-5.4 |
|---|---|---|
| FrontierMath (Tiers 1-3) | 59.2% | 78.6% |
| FrontierMath Tier 4 | 29.3% | 49% |
| MathArena Final-Answer Competitions | 67.6% | 83.1% |
| OTIS Mock AIME 2024-2025 | 86.4% | 97.8% |
| ProofBench | 35% | 56% |
| LMArena Math | 1482 | 1488 |
| FrontierMath (Feb 2025 set) | — | 47.6% |
| FrontierMath Tier 4 (v1) | — | 27.1% |
Knowledge GPT-5.4 leads
GLM-5.2: 57.1 (#40), GPT-5.4: 65.3 (#14)
| Benchmark | GLM-5.2 | GPT-5.4 |
|---|---|---|
| GPQA Diamond | 91.9% | 93.3% |
| SimpleQA Verified | 34.2% | 45.1% |
| LMArena Expert | 1486 | 1507 |
| Humanity's Last Exam | — | 36.2% |
| Vectara Hallucination Rate | — | 7% |
Multimodal Not comparable
GLM-5.2: —, GPT-5.4: 43.7 (#20)
| Benchmark | GLM-5.2 | GPT-5.4 |
|---|---|---|
| LMArena Vision | — | 1303 |
| Blueprint-Bench 2 | — | 27.1% |
| Furniture Assembly | — | 37.5% |
| LMArena Document | — | 1471 |
Multilingual Too close to call
GLM-5.2: 55.8 (#26), GPT-5.4: 56.2 (#23)
| Benchmark | GLM-5.2 | GPT-5.4 |
|---|---|---|
| LMArena Non-English | 1459 | 1465 |
| LMArena Chinese | 1519 | 1519 |
| LMArena French | 1479 | 1493 |
| LMArena German | 1468 | 1472 |
| LMArena Japanese | 1451 | 1485 |
| LMArena Korean | 1445 | 1448 |
| LMArena Russian | 1466 | 1480 |
| LMArena Spanish | 1477 | 1454 |
Instruction Following Too close to call
GLM-5.2: 76.9 (#34), GPT-5.4: 77.1 (#27)
| Benchmark | GLM-5.2 | GPT-5.4 |
|---|---|---|
| LMArena Instruction Following | 1465 | 1469 |
Long Context GPT-5.4 leads
GLM-5.2: 45.3 (#43), GPT-5.4: 50.3 (#8)
| Benchmark | GLM-5.2 | GPT-5.4 |
|---|---|---|
| LMArena Longer Query | 1479 | 1473 |
| CL-bench | — | 27.9% |
| CL-bench Life | — | 21.7% |
Writing & Preference GPT-5.4 leads
GLM-5.2: 70.4 (#21), GPT-5.4: 71.9 (#17)
| Benchmark | GLM-5.2 | GPT-5.4 |
|---|---|---|
| LMArena Text | 1470 | 1469 |
| LMArena Creative Writing | 1462 | 1439 |
| EQ-Bench Creative Writing | 1757 | 1840 |
| EQ-Bench 4 | 1222 | 1272 |
| LMArena Multi-Turn | 1469 | 1482 |
Frequently asked questions
Is GLM-5.2 better than GPT-5.4?
GPT-5.4 is the stronger model overall, scoring 59.4 to 51.1 on the Noometry Index. GLM-5.2 costs 2.6× less per token, which makes it the better buy when GPT-5.4's lead doesn't matter for your workload.
Which is cheaper, GLM-5.2 or GPT-5.4?
GLM-5.2 is cheaper. It lists at $1.40 per million input tokens and $4.40 per million output tokens; GPT-5.4 lists at $2.50 and $15.
Is GLM-5.2 or GPT-5.4 better for coding?
GPT-5.4 scores higher on coding benchmarks: 52.6 versus 51.3 in the Noometry coding category.
Which has the bigger context window?
GPT-5.4 does, with 1.05M tokens against 1M.
How many benchmarks do GLM-5.2 and GPT-5.4 share?
48 benchmarks have published results for both models. GLM-5.2 has 51 scored results on Noometry and GPT-5.4 has 68.