Compare / head-to-head
GLM-5.2vsGPT-5.4
GPT-5.4 leads 6 of 8 shared benchmarks. GLM-5.2 is 3x cheaper per token. GPT-5.4 has the larger context window (1.1M tokens).
Benchmarks from cited public sources; pricing from official pages; status from official provider feeds.
Shared benchmarks
2 – 6
GPT-5.4 leads
Cheaper per token
GLM-5.2
3x cheaper, input + output
Larger context
GPT-5.4
1.1M tokens
Provider uptime (30d)
— · 95.4416%
Z.ai · OpenAI
GLM-5.2
Z.ai · released 2026-06-24
text
Quality
Benchmark matrix
8 shared · 17 only GLM-5.2 · 12 only GPT-5.4| Benchmark | GLM-5.2 | GPT-5.4 | Δ | Edge |
|---|---|---|---|---|
| Reported by both · 8 | ||||
| GPQA Diamond | 91.9% ↗epoch | 92.8% ↗ | -0.9 pt | GPT-5.4 |
| AIME 2026 | 90% ↗matharena ⚠ | 99.2% ↗matharena ⚠ | -9.2 pt | GPT-5.4 |
| FrontierMath Tier 4 v2 (Epoch AI run) | 29.3% ↗epoch | 49% ↗epoch | -19.7 pt | GPT-5.4 |
| FrontierMath Tiers 1-3 v2 (Epoch AI run) | 59.2% ↗epoch | 78.6% ↗epoch | -19.4 pt | GPT-5.4 |
| OTIS Mock AIME 2024-2025 (Epoch AI run) | 86.4% ↗epoch | 97.8% ↗epoch | -11.4 pt | GPT-5.4 |
| SimpleQA Verified | 34.2% ↗epoch | 45.1% ↗epoch | -10.9 pt | GPT-5.4 |
| SWE-Bench Pro | 62.1% ↗ | 57.7% ↗ | +4.4 pt | GLM-5.2 |
| SWE-bench Verified (Epoch AI run) | 78.7% ↗epoch | 76.9% ↗epoch | +1.8 pt | GLM-5.2 |
| Only GLM-5.2 reports · 17 | ||||
| Agents' Last Exam (ALE-CLI) | 23.8% ↗ | not reported | — | — |
| AutomationBench v1.0.6 | 26.2% ↗ | not reported | — | — |
| CyberGym | 77.2% ↗ | not reported | — | — |
| DeepSWE v1.1 | 46.2% ↗ | not reported | — | — |
| ExploitBench | 24.4% ↗ | not reported | — | — |
| ExploitGym 2h | 29 ↗ | not reported | — | — |
| ExploitGym 6h | 39 ↗ | not reported | — | — |
| FrontierSWE | 67.5% ↗ | not reported | — | — |
| GDPval-AA v2 | 1508 ↗ | not reported | — | — |
| HLE (with tools) | 54.7% ↗ | not reported | — | — |
| NL2Repo | 48.9% ↗ | not reported | — | — |
| PostTrainBench | 31.7% ↗ | not reported | — | — |
| ProgramBench (Almost Solved) | 9.5% ↗ | not reported | — | — |
| SWE-Marathon v1.1 | 19.4% ↗ | not reported | — | — |
| Terminal-Bench 2.1 | 81% ↗ | not reported | — | — |
| Terminal-Bench 3.0 | 4.6% ↗ | not reported | — | — |
| Toolathlon Verified | 59.9% ↗ | not reported | — | — |
| Only GPT-5.4 reports · 12 | ||||
| Humanity's Last Exam (no tools) | not reported | 39.8% ↗ | — | — |
| LMArena Elo | not reported | 1452.6 ↗ | — | — |
| BrowseComp | not reported | 82.7% ↗ | — | — |
| GDPval (wins or ties) | not reported | 83% ↗ | — | — |
| Humanity's Last Exam (with tools) | not reported | 52.1% ↗ | — | — |
| MCP Atlas | not reported | 67.2% ↗ | — | — |
| MMMU-Pro (no tools) | not reported | 81.2% ↗ | — | — |
| MMMU-Pro (with tools) | not reported | 82.1% ↗ | — | — |
| OSWorld-Verified | not reported | 75% ↗ | — | — |
| Tau2-bench Telecom | not reported | 98.9% ↗ | — | — |
| Terminal-Bench 2.0 | not reported | 75.1% ↗ | — | — |
| Toolathlon | not reported | 54.6% ↗ | — | — |
Scores tagged "epoch" or "matharena" are independent runs, used only where the lab has not published its own; ⚠ marks rows MathArena flags as released after the competition. Higher is better on every row. Δ is GLM-5.2 minus GPT-5.4 in the benchmark's own unit. "Not reported" means the lab has not published that figure; it is not a zero. ↗ opens the source.
Specs & pricing
Side by side
Official pricing pages and model cards| Spec | GLM-5.2 | GPT-5.4 | Edge |
|---|---|---|---|
| Context window | 1M tokens | 1.1M tokens | GPT-5.4 |
| Max output | 131K tokens | 128K tokens | GLM-5.2 |
| Input price / 1M | $1.4 ↗ | $2.5 ↗ | GLM-5.2 |
| Output price / 1M | $4.4 ↗ | $15 ↗ | GLM-5.2 |
| Cached input / 1M | $0.26 ↗ | $0.25 ↗ | GPT-5.4 |
| Input + output / 1M Lower is cheaper. List prices; batch, tool and regional fees excluded. | $5.8 | $17.5 | GLM-5.2 |
| Modalities | text | text · vision | GPT-5.4 |
| Released | 2026-06-24 | 2026-03-05 | — |
| Cited benchmark scores | 26 | 21 | — |
Reliability
Provider status
Last 60 days · refreshed daily on this page · live on /statusMore matchups
GLM-5.2 vs …
Models sharing the most benchmarksMore matchups
GPT-5.4 vs …
Models sharing the most benchmarksBuilt by Respan
Which one wins on your data?
Public benchmarks are a starting point. Run GLM-5.2 and GPT-5.4 on your own prompts with Respan evals, or route to either through one gateway key with automatic failover.