Model comparison
GLM-5 vs GPT-5.4
GPT-5.4 is the stronger model overall, scoring 59.4 to 46.1 on the Noometry Index. GLM-5 costs 3.6× less per token, which makes it the better buy when GPT-5.4's lead doesn't matter for your workload.
Last verified . 39 shared benchmarks.
Summary
- They share 39 benchmarks with published results for both. GLM-5 scores higher in 0 categories and GPT-5.4 in 9 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where GPT-5.4 leads 61.8 to 27.6.
- The biggest single-benchmark swing is ARC-AGI-2: 4.9% for GLM-5 and 74% for GPT-5.4.
- GLM-5 is cheaper at $1 / $3.20 per million input/output tokens, against $2.50 / $15 for GPT-5.4.
- GPT-5.4 accepts more context: 1.05M tokens versus 205K.
- GLM-5 has downloadable open weights; the other is API-only.
Side by side
| GLM-5 | GPT-5.4 | |
|---|---|---|
| Provider | Z.ai (Zhipu) | OpenAI |
| Noometry Index | 46.1 | 59.4 |
| Released | 2026-02-11 | 2026-03-05 |
| Weights | Open | Proprietary |
| Context window | 205K | 1.05M |
| Max output | 131K | 128K |
| Input $ / M tokens | $1 | $2.50 |
| Output $ / M tokens | $3.20 | $15 |
| Results tracked | 45 | 68 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-5.4 leads
GLM-5: 49.0 (#52), GPT-5.4: 52.6 (#33)
| Benchmark | GLM-5 | GPT-5.4 |
|---|---|---|
| SWE-bench Verified | 72.1% | 76.9% |
| LMArena WebDev | 1434 | 1465 |
| WeirdML | 48.2% | 77.7% |
| LMArena Coding | 1461 | 1497 |
| ALE-Bench | 765.62 | 1,607 |
| DeepSWE | — | 51.8% |
| SWE-bench Verified (bash only) | 72.8% | — |
| SWE-bench Multilingual | 69.7% | — |
| SciCode | — | 56.6% |
| GSO | — | 31.4% |
| MirrorCode | — | 15.6% |
| AlgoTune | — | 1.85 |
Agentic & Tool Use GPT-5.4 leads
GLM-5: 31.1 (#71), GPT-5.4: 46.5 (#13)
| Benchmark | GLM-5 | GPT-5.4 |
|---|---|---|
| Terminal-Bench | 52.4% | 81.8% |
| τ²-bench Banking | 9.8% | 39.4% |
| Vending-Bench 2 | 4,432 | 6,144 |
| APEX-Agents | — | 52.4% |
| τ²-bench Airline | 82.5% | — |
| τ²-bench Retail | 73.7% | — |
| τ²-bench Telecom | 86.8% | — |
| DeepResearch Bench | — | 35.1% |
| PostTrainBench | — | 19% |
| GBAEval | — | 45.1% |
| LMArena Search | — | 1197 |
| METR Time Horizons | — | 74.3% |
Reasoning GPT-5.4 leads
GLM-5: 27.6 (#116), GPT-5.4: 61.8 (#19)
| Benchmark | GLM-5 | GPT-5.4 |
|---|---|---|
| ARC-AGI-2 | 4.9% | 74% |
| Kagi LLM Benchmark | 75% | 63.8% |
| NYT Connections (extended) | 74.8% | 91.3% |
| ARC-AGI-1 | 44.7% | 93.7% |
| Chess Puzzles | 10% | 44% |
| LMArena Hard Prompts | 1452 | 1485 |
| Epoch Capabilities Index | 145.83 | 156.81 |
| ForecastBench | 61 | 59.5 |
| SimpleBench | 53.2% | — |
| CritPt | — | 23.4% |
| EnigmaEval | — | 16% |
| Thematic Generalization | — | 80% |
| EBR-Bench | — | 25.4% |
| Mystery Game Puzzles | — | 37% |
| DTBench | — | 94.4% |
| LMCA | — | 52% |
Math GPT-5.4 leads
GLM-5: 46.4 (#71), GPT-5.4: 73.5 (#19)
| Benchmark | GLM-5 | GPT-5.4 |
|---|---|---|
| MathArena Final-Answer Competitions | 65.7% | 83.1% |
| OTIS Mock AIME 2024-2025 | 80% | 97.8% |
| LMArena Math | 1440 | 1488 |
| FrontierMath (Feb 2025 set) | 16.4% | 47.6% |
| FrontierMath Tier 4 (v1) | 2.1% | 27.1% |
| FrontierMath (Tiers 1-3) | — | 78.6% |
| FrontierMath Tier 4 | — | 49% |
| ProofBench | — | 56% |
Knowledge GPT-5.4 leads
GLM-5: 52.3 (#64), GPT-5.4: 65.3 (#14)
| Benchmark | GLM-5 | GPT-5.4 |
|---|---|---|
| GPQA Diamond | 87.8% | 93.3% |
| Vectara Hallucination Rate | 10.1% | 7% |
| LMArena Expert | 1454 | 1507 |
| Humanity's Last Exam | — | 36.2% |
| SimpleQA Verified | — | 45.1% |
Multimodal Not comparable
GLM-5: —, GPT-5.4: 43.7 (#20)
| Benchmark | GLM-5 | GPT-5.4 |
|---|---|---|
| LMArena Vision | — | 1303 |
| Blueprint-Bench 2 | — | 27.1% |
| Furniture Assembly | — | 37.5% |
| LMArena Document | — | 1471 |
Multilingual GPT-5.4 leads
GLM-5: 53.7 (#58), GPT-5.4: 56.2 (#23)
| Benchmark | GLM-5 | GPT-5.4 |
|---|---|---|
| LMArena Non-English | 1430 | 1465 |
| LMArena Chinese | 1511 | 1519 |
| LMArena French | 1455 | 1493 |
| LMArena German | 1445 | 1472 |
| LMArena Japanese | 1416 | 1485 |
| LMArena Korean | 1423 | 1448 |
| LMArena Russian | 1436 | 1480 |
| LMArena Spanish | 1454 | 1454 |
Instruction Following GPT-5.4 leads
GLM-5: 75.2 (#67), GPT-5.4: 77.1 (#27)
| Benchmark | GLM-5 | GPT-5.4 |
|---|---|---|
| LMArena Instruction Following | 1428 | 1469 |
Long Context GPT-5.4 leads
GLM-5: 44.7 (#60), GPT-5.4: 50.3 (#8)
| Benchmark | GLM-5 | GPT-5.4 |
|---|---|---|
| CL-bench | 18.7% | 27.9% |
| LMArena Longer Query | 1446 | 1473 |
| CL-bench Life | — | 21.7% |
Writing & Preference GPT-5.4 leads
GLM-5: 66.0 (#38), GPT-5.4: 71.9 (#17)
| Benchmark | GLM-5 | GPT-5.4 |
|---|---|---|
| LMArena Text | 1446 | 1469 |
| LMArena Creative Writing | 1439 | 1439 |
| EQ-Bench Creative Writing | 1601 | 1840 |
| LMArena Multi-Turn | 1456 | 1482 |
| EQ-Bench 4 | — | 1272 |
Frequently asked questions
Is GLM-5 better than GPT-5.4?
GPT-5.4 is the stronger model overall, scoring 59.4 to 46.1 on the Noometry Index. GLM-5 costs 3.6× 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 or GPT-5.4?
GLM-5 is cheaper. It lists at $1 per million input tokens and $3.20 per million output tokens; GPT-5.4 lists at $2.50 and $15.
Is GLM-5 or GPT-5.4 better for coding?
GPT-5.4 scores higher on coding benchmarks: 52.6 versus 49.0 in the Noometry coding category.
Which has the bigger context window?
GPT-5.4 does, with 1.05M tokens against 205K.
How many benchmarks do GLM-5 and GPT-5.4 share?
39 benchmarks have published results for both models. GLM-5 has 45 scored results on Noometry and GPT-5.4 has 68.