Model comparison
DeepSeek-V3 vs GLM-4.5
GLM-4.5 is the stronger model overall, scoring 42.0 to 39.5 on the Noometry Index. DeepSeek-V3 costs 2.5× less per token, which makes it the better buy when GLM-4.5's lead doesn't matter for your workload.
Last verified . 23 shared benchmarks.
DeepSeekSummary
- They share 23 benchmarks with published results for both. DeepSeek-V3 scores higher in 2 categories and GLM-4.5 in 6 categories; 6 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where GLM-4.5 leads 28.6 to 20.5.
- The biggest single-benchmark swing is Confabulations: 26.1% for DeepSeek-V3 and 11.3% for GLM-4.5.
- DeepSeek-V3 is cheaper at $0.24 / $0.90 per million input/output tokens, against $0.60 / $2.20 for GLM-4.5.
- DeepSeek-V3 accepts more context: 164K tokens versus 131K.
Side by side
| DeepSeek-V3 | GLM-4.5 | |
|---|---|---|
| Provider | DeepSeek | Z.ai (Zhipu) |
| Noometry Index | 39.5 | 42.0 |
| Released | 2024-12-26 | 2025-07-27 |
| Weights | Open | Open |
| Context window | 164K | 131K |
| Max output | 164K | 98K |
| Input $ / M tokens | $0.24 | $0.60 |
| Output $ / M tokens | $0.90 | $2.20 |
| Results tracked | 60 | 27 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Too close to call
DeepSeek-V3: 42.3 (#106), GLM-4.5: 41.4 (#125)
| Benchmark | DeepSeek-V3 | GLM-4.5 |
|---|---|---|
| WeirdML | 36.1% | 40.6% |
| LMArena Coding | 1368 | 1434 |
| SWE-bench Verified (bash only) | — | 54.2% |
| Aider Polyglot | 55.1% | — |
| SciCode | 35.8% | — |
| BigCodeBench Instruct | 50% | — |
| LiveBench Coding | 70.9% | — |
| BigCodeBench Complete | 62.2% | — |
| ALE-Bench | — | 344.82 |
| AlgoTune | — | 1.52 |
| HumanEval+ | 86.6% | — |
| MBPP+ | 73% | — |
Agentic & Tool Use Not comparable
DeepSeek-V3: —, GLM-4.5: —
| Benchmark | DeepSeek-V3 | GLM-4.5 |
|---|---|---|
| METR Time Horizons | 49.6% | — |
Reasoning GLM-4.5 leads
DeepSeek-V3: 20.5 (#236), GLM-4.5: 28.6 (#100)
| Benchmark | DeepSeek-V3 | GLM-4.5 |
|---|---|---|
| Kagi LLM Benchmark | 52.3% | 57.9% |
| LMArena Hard Prompts | 1365 | 1429 |
| SimpleBench | 27.2% | — |
| CritPt | 0% | — |
| LiveBench Reasoning | 65.8% | — |
| DTBench | 64.8% | — |
| LiveBench Data Analysis | 60.9% | — |
| LMCA | 15.5% | — |
| BIG-Bench Hard | 87.5% | — |
| Epoch Capabilities Index | 135.94 | — |
| ForecastBench | 59.1 | — |
| HellaSwag | 88.9% | — |
| LiveBench | 66.9% | — |
| PIQA | 84.7% | — |
| WinoGrande | 85.2% | — |
Math GLM-4.5 leads
DeepSeek-V3: 32.1 (#219), GLM-4.5: 39.0 (#116)
| Benchmark | DeepSeek-V3 | GLM-4.5 |
|---|---|---|
| LMArena Math | 1373 | 1427 |
| OTIS Mock AIME 2024-2025 | 37.8% | — |
| Omni-MATH | 40.3% | — |
| LiveBench Math | 73.5% | — |
| MATH Level 5 | 75.5% | — |
| FrontierMath (Feb 2025 set) | 1.7% | — |
Knowledge DeepSeek-V3 leads
DeepSeek-V3: 37.5 (#155), GLM-4.5: 35.9 (#179)
| Benchmark | DeepSeek-V3 | GLM-4.5 |
|---|---|---|
| Confabulations | 26.1% | 11.3% |
| LMArena Expert | 1351 | 1433 |
| GPQA Diamond | 67.6% | — |
| Humanity's Last Exam | — | 8.3% |
| MMLU-Pro | 72.3% | — |
| Vectara Hallucination Rate | 6.1% | — |
| GPQA (HELM) | 53.8% | — |
| ARC (AI2) Challenge | 95.3% | — |
| MMLU | 87.2% | — |
| TriviaQA | 82.9% | — |
Multilingual GLM-4.5 leads
DeepSeek-V3: 48.5 (#143), GLM-4.5: 52.8 (#77)
| Benchmark | DeepSeek-V3 | GLM-4.5 |
|---|---|---|
| LMArena Non-English | 1358 | 1417 |
| LMArena Chinese | 1391 | 1465 |
| LMArena French | 1385 | 1418 |
| LMArena German | 1374 | 1407 |
| LMArena Japanese | 1333 | 1415 |
| LMArena Korean | 1319 | 1380 |
| LMArena Russian | 1373 | 1414 |
| LMArena Spanish | 1358 | 1454 |
Instruction Following GLM-4.5 leads
DeepSeek-V3: 72.8 (#130), GLM-4.5: 74.1 (#104)
| Benchmark | DeepSeek-V3 | GLM-4.5 |
|---|---|---|
| LMArena Instruction Following | 1345 | 1404 |
| LiveBench Instruction Following | 81.5% | — |
| IFEval | 83.2% | — |
Long Context GLM-4.5 leads
DeepSeek-V3: 34.0 (#253), GLM-4.5: 38.2 (#201)
| Benchmark | DeepSeek-V3 | GLM-4.5 |
|---|---|---|
| Fiction.LiveBench | 50% | 58.3% |
| LMArena Longer Query | 1352 | 1412 |
Writing & Preference Too close to call
DeepSeek-V3: 57.4 (#130), GLM-4.5: 57.5 (#127)
| Benchmark | DeepSeek-V3 | GLM-4.5 |
|---|---|---|
| LMArena Text | 1375 | 1430 |
| LMArena Creative Writing | 1364 | 1395 |
| Short-Story Creative Writing | 77% | 73.4% |
| EQ-Bench Creative Writing | 1472 | 1343 |
| LMArena Multi-Turn | 1389 | 1415 |
| WildBench | 83% | — |
| LiveBench Language | 49.1% | — |
Frequently asked questions
Is DeepSeek-V3 better than GLM-4.5?
GLM-4.5 is the stronger model overall, scoring 42.0 to 39.5 on the Noometry Index. DeepSeek-V3 costs 2.5× less per token, which makes it the better buy when GLM-4.5's lead doesn't matter for your workload.
Which is cheaper, DeepSeek-V3 or GLM-4.5?
DeepSeek-V3 is cheaper. It lists at $0.24 per million input tokens and $0.90 per million output tokens; GLM-4.5 lists at $0.60 and $2.20.
Is DeepSeek-V3 or GLM-4.5 better for coding?
They score almost the same on coding (42.3 vs 41.4); test both on your own repository before choosing.
Which has the bigger context window?
DeepSeek-V3 does, with 164K tokens against 131K.
How many benchmarks do DeepSeek-V3 and GLM-4.5 share?
23 benchmarks have published results for both models. DeepSeek-V3 has 60 scored results on Noometry and GLM-4.5 has 27.